{ "api_version": "1.0", "description": "Projects that publish a use case.", "source": "https://fair-forward.github.io/datasets/", "documentation": "https://fair-forward.github.io/datasets/api/v1/index.json", "license": { "spdx": "CC0-1.0", "url": "https://creativecommons.org/publicdomain/zero/1.0/", "note": "Covers the catalogue metadata in this file: you may republish these records freely, with or without attribution. It does NOT cover the linked assets. Each record's `license` field describes the terms of the asset itself, which Fair Forward does not own; a null value means no license has been recorded, not that the asset is unlicensed or free to reuse." }, "version": "465e9286b50af245", "count": 38, "vocabularies": { "sdgs": [ "SDG 2", "SDG 4", "SDG 5", "SDG 7", "SDG 8", "SDG 9", "SDG 10", "SDG 11", "SDG 12", "SDG 13", "SDG 14", "SDG 15", "SDG 16" ], "data_types": [ "Drone Imagery", "Geospatial/Remote Sensing", "Images", "Meterological", "Other", "Tabular", "Text", "Voice" ], "countries": [ { "name": "Angola", "iso2": "AO" }, { "name": "Benin", "iso2": "BJ" }, { "name": "Brazil", "iso2": "BR" }, { "name": "Cameroon", "iso2": "CM" }, { "name": "Colombia", "iso2": "CO" }, { "name": "Cote d'Ivoire", "iso2": "CI" }, { "name": "Democratic Republic of Congo", "iso2": "CD" }, { "name": "Ecuador", "iso2": "EC" }, { "name": "Ethiopia", "iso2": "ET" }, { "name": "Ghana", "iso2": "GH" }, { "name": "India", "iso2": "IN" }, { "name": "Indonesia", "iso2": "ID" }, { "name": "Kenya", "iso2": "KE" }, { "name": "Madagascar", "iso2": "MG" }, { "name": "Mozambique", "iso2": "MZ" }, { "name": "Niger", "iso2": "NE" }, { "name": "Nigeria", "iso2": "NG" }, { "name": "Pakistan", "iso2": "PK" }, { "name": "Rwanda", "iso2": "RW" }, { "name": "Senegal", "iso2": "SN" }, { "name": "South Africa", "iso2": "ZA" }, { "name": "Tanzania", "iso2": "TZ" }, { "name": "Togo", "iso2": "TG" }, { "name": "Uganda", "iso2": "UG" }, { "name": "Zambia", "iso2": "ZM" } ], "regions": [ "East Africa", "Global", "West Africa" ], "maturity_stages": [ "dataset", "model", "pilot", "usecase", "business" ] }, "projects": [ { "id": "ui_8", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_8-using_locallanguage_ai_advise_women_in/", "aliases": [ "lifting_up_women_through_land_ownership", "using_locallanguage_ai_advise_women_in" ], "title": "Using local-language AI advise women in DRC on land ownership - Haki des femmes", "description": { "text": "Haki will leverage voice technology to provide access to legal information and support for women in Katanga and Lualaba provinces of the Democratic Republic of Congo to ensure they have the right to access, use, inherit, control, and own land. Majority of women in DRC often lose their access to land after the passing of a loved one or husband due to lack of knowledge of land rights. This solution will help women to access information and legal support in securing their land rights in Kiswahili. This was part of a Mozilla innovation challenge supporting people and projects across East Africa who leverage Common Voice’s open-source voice data set to unlock social and economic opportunities. These grants help to advance the use of open-source voice data for products that support community participation and engagement.", "provenance": "curated" }, "kind": [ "usecase" ], "countries": [ { "name": "Democratic Republic of Congo", "iso2": "CD" } ], "regions": [], "sdgs": [ "SDG 5", "SDG 2", "SDG 10" ], "data_types": [ "Text", "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Core23Lab", "links": [] }, "catalyzed_by": { "text": "Mozilla Foundation & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "Gates Foundation & BMZ", "links": [] } }, "contact": "Core23Lab (engage@core23lab.org)", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "play.google.com: details", "url": "https://play.google.com/store/apps/details?id=org.core23lab.hdf&pcampaignid=web_share&pli=1" } ], "additional": [ { "label": "Lifting Up Women Through Land Ownership (mozillafoundation.org)", "url": "https://www.mozillafoundation.org/de/blog/lifting-up-women-through-land-ownership/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Haki des Femmes is an Android app by Core23lab providing legal information on land ownership rights for women in the Katanga and Lualaba provinces of DRC. Content is in Kiswahili. Covers rights to access, use, inherit, control, and own land. The app uses voice technology for interaction. Data safety: does not share data with third parties; personal data encrypted in transit; users can request data deletion. Available for all ages on Google Play.\n\nSource: https://play.google.com/store/apps/details?id=org.core23lab.hdf", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Haki des Femmes is a voice-enabled Android application by Core23lab. Users speak queries in Kiswahili about land ownership rights, and the app returns relevant legal information and guidance. Designed for women in the Katanga and Lualaba provinces of DRC who risk losing land access after the death of a family member. Developer contact: devs.core23lab@gmail.com.\n\nSource: https://play.google.com/store/apps/details?id=org.core23lab.hdf", "provenance": "auto-enriched" }, "how_to_use": { "text": "Haki des Femmes is a ready-to-use voice-enabled chatbot app that provides women in the Democratic Republic of Congo with accessible legal information about land ownership rights. It operates in Congolese Swahili (Kiswahili) and can be downloaded directly from the Google Play Store by searching for \"Haki des femmes.\"\n\nThe app is designed for community organizations, legal aid providers, and development practitioners working on women's land rights in the DRC's Katanga and Lualaba provinces. Many women in these communities are unaware of existing laws that allow them to own land, face barriers to proper documentation, or lack legal marriages that would confer inheritance rights. The chatbot simplifies this legal information through a voice interface, helping women understand the concrete steps needed to secure land ownership -- including the process of legalizing marriages as a prerequisite for land rights.\n\nDevelopment practitioners can use Haki des Femmes as a model for building similar legal information tools in other contexts. The approach of combining voice technology with local-language legal guidance could be adapted for other jurisdictions or legal domains where access to legal literacy is a barrier.\n\nHaki des Femmes was developed by Core23Lab as part of Mozilla's 2023-24 Common Voice Kiswahili program, which funds projects using Kiswahili voice technology to support marginalized groups in Kenya, Tanzania, and the DRC. The team conducted surveys of women in Katanga and Lualaba provinces to identify specific knowledge gaps about land ownership before designing the chatbot, an approach worth replicating in similar projects. More background on the project rationale is available on the Mozilla Foundation blog.\n\nKnown limitations: The app is specific to DRC land law and Congolese Swahili and is not directly applicable to other countries or legal systems. Voice-based interaction requires a smartphone with a microphone and internet access.\n\nSources:\n- https://play.google.com/store/apps/details?id=org.core23lab.hdf\n- https://www.mozillafoundation.org/en/blog/lifting-up-women-through-land-ownership/", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_8/images/placeholder_image.jpeg" }, { "id": "ui_11", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_11-indigenous_knowledge_meets_ai_ethical_monitoring/", "aliases": [ "indigenous_knowledge_meets_ai_ethical_monitoring" ], "title": "Indigenous Knowledge Meets AI: Ethical monitoring of climate stress and biodiversity: sounds of elephants and Katip (Ltome-Katip) in Kenya and the Ecuadorian Amazon", "description": { "text": "The Ltome-Katip datasets are the first Indigenous-labelled bioacoustic datasets designed specifically to support the development of ethical AI for biodiversity monitoring. Co-created by Indigenous data stewards from the Samburu tribe in northern Kenya and the Shuar Nation in the Ecuadorian Amazon, the recordings focus on two sentinel species: Ltome (elephant) and Katip (rodent), both of which signal ecological shifts under climate stress. All data were collected, annotated, and governed by the Indigenous communities, following locally defined protocols rooted in Indigenous data sovereignty. These datasets are not only scientifically valuable — they establish a precedent for how Indigenous communities can lead in setting standards for responsible, consent-based AI development.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Ecuador", "iso2": "EC" }, { "name": "Kenya", "iso2": "KE" } ], "regions": [], "sdgs": [ "SDG 15" ], "data_types": [ "Geospatial/Remote Sensing", "Other" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Space4Innovation, Namunyak Conservancy, GEO Indigenous Alliance, MUSAP & Rochester Institute for Technology", "links": [] }, "catalyzed_by": { "text": "Lacuna-Fund / Meridian (Climate-call) & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Space4Innovation, Diana Mastracci (diana@space4innovation.com)", "access_note": null, "links": { "dataset": [ { "label": "space4innovation.github.io: index", "url": "https://space4innovation.github.io/ltomekatip/index.html" } ], "usecase": [ { "label": "arbimon.org: namunyak-conservancy-reteti-eleph…", "url": "https://arbimon.org/p/namunyak-conservancy-reteti-elephant-sanctuary" }, { "label": "arbimon.org: insights", "url": "https://arbimon.org/p/shakiam-ecuadorian-amazon/insights" } ], "additional": [ { "label": "When Future Already Here Now Its Being Filmed Diana Mastracci Sanchez Y5Rue (linkedin.com)", "url": "https://www.linkedin.com/pulse/when-future-already-here-now-its-being-filmed-diana-mastracci-sanchez-y5rue" }, { "label": "Embracing Uncertainty Hidden Strength Science Diana Mastracci Sanchez Qdj5E (linkedin.com)", "url": "https://www.linkedin.com/pulse/embracing-uncertainty-hidden-strength-science-diana-mastracci-sanchez-qdj5e" }, { "label": "Bridging Worlds Indigenous Led Innovation Remote Mastracci Sanchez Yjqfe (linkedin.com)", "url": "https://www.linkedin.com/pulse/bridging-worlds-indigenous-led-innovation-remote-mastracci-sanchez-yjqfe" }, { "label": "Ltome Katip Indigenous Led Labelling Inclusive Ai Addressing Human Wildlife Conflict And (rit.edu)", "url": "https://www.rit.edu/dirs/research/ltome-katip-indigenous-led-labelling-inclusive-ai-addressing-human-wildlife-conflict-and" }, { "label": "Professor Helps Bring Machine Learning Indigenous Communities (rit.edu)", "url": "https://www.rit.edu/news/professor-helps-bring-machine-learning-indigenous-communities" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Ltome-Katip Indigenous Bioacoustic Dataset\nRegions: Samburu (Kenya) · Shuar (Ecuadorian Amazon)\nCustodians: Chief Titus Letaapo (Samburu tribe) (Namunyak Conservancy), Chief Mario Vargas Shakaim (Shuar Nation) (MUSAP Biological Station), and Space4Innovation\nThis dataset contains Indigenous-labelled bioacoustic recordings from two ecosystems—semi-arid savannah and tropical rainforest—collected through AudioMoth bioacustic sensors. Data include species-specific sounds (e.g., elephants, rodents), environmental background, and associated metadata following the CARE Principles for Indigenous Data Governance.\nUse cases: biodiversity monitoring, species classification, human–wildlife conflict alerts, and AI model training for conservation.\nLimitations: class imbalance (key species overrepresented), environmental noise, and spatial clustering; users should apply noise filtering and ethical review before reuse. These audio data are collected 24/7 when deployed in time periods ranging from hours to several weeks. The data are acquired from multiple microphones spread across the study site. Each microphone has a unique serial number and the geographic locations are provided using GPS. The data are time stamped, however there are data gaps in time and space due to logistics, equipment failure, or power loss. The original data are stored as 16-bit WAV files and are available. To make the data more widely available, they have been uploaded to the Arbimon.org platform. The Arbimon cloud platform is built for bioacoustics analysis using various ML .", "provenance": "curated" }, "model_characteristics": { "text": "The Ltome-Katip system uses Indigenous-labelled bioacoustic data to train AI models that detect and classify species like elephants in Samburu (Kenya) and rodents in the Ecuadorian Amazon. These models are already being used to monitor biodiversity, understand ecological stress, and support early warning systems rooted in Indigenous governance. What sets Ltome-Katip apart is that both the dataset and its governance model were co-designed by Indigenous communities. All development follows the CARE Principles for Indigenous Data Sovereignty, and any replication must go through an Ethical AI Assessment to ensure consent, transparency, and benefit-sharing. This project sets a new global benchmark for community-led, responsible AI in biodiversity and conservation.\n\nThe data is processed using the Arbimon platform, where recordings are visualized as spectrograms and labelled through bounding boxes by trained Indigenous data stewards. The outputs — including geospatial and temporal metadata — can be downloaded as CSV files and used for further machine learning or integration with other ecological datasets. These tools are already generating insights into ecosystem change and human–wildlife conflict. The core team is now actively designing Ltome-Katip 2, a next-phase expansion that will deepen Indigenous-led data infrastructure, extend sensor coverage, and explore AI integration with the Namunyak app. While plans are in development, we are currently seeking aligned funding to support this work, which will remain entirely Indigenous-led and ethically governed at every stage.", "provenance": "curated" }, "how_to_use": { "text": "The Ltome-Katip datasets can already be used to detect and classify species such as elephants and rodents, enabling real-time biodiversity monitoring and alerts for human–wildlife conflict. They also support ecosystem health assessments by capturing patterns in species richness, activity cycles, and climate-driven changes. Indigenous-led early warning systems are already being built using these datasets and dashboards, allowing communities to visualize and act on local ecological shifts. Researchers can extend this work by adding new species, integrating satellite data, applying transfer learning, or developing explainable AI tools to improve accuracy and cross-ecosystem usability. All reuse must respect Indigenous data sovereignty, undergo an ethical AI review, and credit the original communities. The Ltome-Katip core team is actively seeking funding for the next phase — Ltome-Katip 2 — which will expand the sensor network, strengthen community data infrastructure, and integrate AI capabilities into the Namunyak Indigenous app.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_11/images/placeholder_image.jpeg" }, { "id": "ui_12", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_12-aipowered_detection_of_diseases_for_cashew/", "aliases": [ "aipowered_detection_of_diseases_for_cashew", "cadi_ai_project_ml4cashew", "drone_images_of_disease_manifestations_in" ], "title": "AI-powered detection of diseases for Cashew farmers in Ghana", "description": { "text": "Imagine, that you are a small-holder farmer in Ghana fearing crop disease in your Cashew farm. You also know that early intervention could increase yields by up to 30% - The Cashew Disease Identification (CADI AI) dataset and application is there to make early detection of diseases in cashew plantations in Ghana through AI possible. This helps securing livelihoods, boosting food security, and fueling further economic growth. You will be able to use an openly accessible data set (4,736 UAV images), a machine learning model, and a desktop app to replicate this approach.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Ghana", "iso2": "GH" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 13" ], "data_types": [ "Drone Imagery" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "AGPL 3.0", "spdx": "AGPL-3.0", "url": null }, "organizations": { "provided_by": { "text": "Kara Agro", "links": [ { "name": "Kara Agro", "url": "https://karaagro.com/" } ] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "KaraAgro (darlington@gudra-studio.com)", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: CADI-AI", "url": "https://huggingface.co/datasets/KaraAgroAI/CADI-AI" } ], "usecase": [ { "label": "Hugging Face: CADI-AI", "url": "https://huggingface.co/KaraAgroAI/CADI-AI" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "A responsible AI Assessment was undertaken for this dataset / use case to help AI developers and project managers to identify, assess and mitigate potential harms and biases in AI. For methodology, see https://www.bmz-digital.global/en/news/ethical-crash-test-for-ai-how-to-navigate-the-road-to-responsible-innovation/ \n\nLicense: https://www.gnu.org/licenses/agpl-3.0.html", "provenance": "curated" }, "model_characteristics": { "text": "CADI-AI (Cashew Disease Identification with AI) by KaraAgro AI Foundation, funded by GIZ through MOVE and FAIR Forward initiatives on behalf of BMZ. Model: YOLOv5x object detection, trained on 3,788 drone-captured images at 640x640 input resolution. Detects 3 classes: insect damage, disease (microbial), and abiotic stress. Performance (mAP@50): 0.648 overall, 0.815 insect, 0.588 disease, 0.542 abiotic. Dataset: 4,736 images total (train/val/test) with 22,610 annotated bounding boxes in YOLO format. Dataset license: CC BY-SA 4.0. Model license: AGPL-3.0. Demo available on Hugging Face Spaces; desktop app on GitHub (karaagro/cadi-ai).\n\nSource: https://huggingface.co/datasets/KaraAgroAI/CADI-AI, https://huggingface.co/KaraAgroAI/CADI-AI", "provenance": "auto-enriched" }, "how_to_use": { "text": "The CADI-AI project is useful for anyone working on cashew crop health monitoring, agricultural extension, or precision agriculture in West Africa. It provides both a labeled image dataset and a ready-to-use pre-trained model for detecting three types of cashew tree health issues -- abiotic stress, disease damage, and insect damage -- from drone-captured imagery.\n\nIf you want to try the model immediately, a live demo is available on HuggingFace Spaces where you can upload your own cashew tree images and see detection results without any setup. For deployment in the field, a desktop application is also available on GitHub. These tools allow agricultural extension officers and agronomists to identify health issues across cashew plantations quickly, enabling targeted interventions rather than blanket treatments.\n\nThe dataset itself contains 4,736 high-resolution drone images (1600x1300 pixels) with over 22,000 annotated instances across the three health-issue classes, licensed under CC-BY-SA 4.0. Researchers and developers can use this data to train improved detection models or to extend the approach to other tree crops. The annotations are in YOLO format, and the pre-trained YOLOv5x model achieves a mean average precision (mAP@50) of 0.65, with strongest performance on insect damage detection (0.82 mAP@50) due to its distinct visual features. Disease and abiotic stress classes are harder to distinguish because their symptoms can overlap in field conditions -- an area where further research could improve accuracy.\n\nA detailed datasheet documenting the data collection methodology is available via the HuggingFace dataset card. The dataset (approximately 3.78 GB) and model (approximately 173 MB) can be downloaded from HuggingFace after acknowledging the license terms.\n\nCost and resources: The dataset and model are freely available. Deploying the model requires only standard compute resources. The CADI-AI project was created by the KaraAgro AI Foundation, funded by GIZ and BMZ through the FAIR Forward and MOVE programs.\n\nSources:\n- https://huggingface.co/datasets/KaraAgroAI/CADI-AI\n- https://huggingface.co/KaraAgroAI/CADI-AI", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_12/images/Screenshot 2025-05-13 at 11.58.32.png" }, { "id": "ui_13", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_13-supporting_food_security_and_climate_change/", "aliases": [ "ghana_biomass_challenge_ghana_crop_disease", "supporting_food_security_and_climate_change" ], "title": "Supporting food security and climate change adaptation: AI-powered crop disease identification for maize, tomatoes, pepper in Ghana", "description": { "text": "This dataset helps to build and improve crop disease detection systems for smallholder farming in Ghana and similar West African contexts. It covers three crops -- tomatoes, pepper and maize -- with 22 disease and health classes in total, making it one of the more comprehensive Afrocentric crop disease image collections available. The data was collected in 10 districts of the Ashanti Region of Ghana by the RAIL-KNUST team and the Plant Protection and Research Services Directorate (PPRSD) of the Ministry of Food and Agriculture. A 3-month long data challenge was hosted on Zindi based on this data set and the 3 winning models are also available as open source resources to serve as your base models in your research into crop diseases in Ghana.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Ghana", "iso2": "GH" } ], "regions": [], "sdgs": [ "SDG 2" ], "data_types": [ "Images" ], "maturity": { "stage": "Dataset > Model", "tags": [ "dataset", "model" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Kwame Nkrumah University of Science and Technology (KNUST), Responsible AI Lab, Plant Protection and Regulatory Services Directorate (Ghana)", "links": [ { "name": "Responsible AI Lab", "url": "https://rail.knust.edu.gh/" } ] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ, Digital Transformation Centre Ghana, GIZ", "links": [ { "name": "GIZ", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Responsible AI Lab (RAIL) at Kwame Nkrumah University of Science and Technology (rail@knust.edu.gh)", "access_note": null, "links": { "dataset": [ { "label": "Kaggle: crop-disease-ghana", "url": "https://www.kaggle.com/datasets/responsibleailab/crop-disease-ghana" } ], "usecase": [ { "label": "Kaggle: code", "url": "https://www.kaggle.com/datasets/responsibleailab/crop-disease-ghana/code" } ], "additional": [ { "label": "Ghana Crop Disease Detection Challenge (zindi.africa)", "url": "https://zindi.africa/competitions/ghana-crop-disease-detection-challenge" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Afrocentric crop disease dataset by Responsible AI Lab. Contains annotated leaf images showing healthy specimens and disease-affected leaves at various crop development phases. Data Type: Image. Size: ~20 GB. License: CC BY 4.0. Version 16 (last modified March 2025). Created with a focus on African agricultural diversity.\n\nSource: https://www.kaggle.com/datasets/responsibleailab/crop-disease-ghana\n\nAlso see here for the Zindi challenge: https://zindi.africa/competitions/ghana-crop-disease-detection-challenge", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Crop disease identification application using computer vision and deep learning on annotated leaf images from African crops. Input: leaf images. Output: disease detection and classification. Dataset created by Responsible AI Lab in collaboration with the Plant Protection and Research Services Directorate (PPRSD) of Ghana's Ministry of Food and Agriculture. Dataset openly available under CC BY 4.0.\n\nSource: https://www.kaggle.com/datasets/responsibleailab/crop-disease-ghana", "provenance": "auto-enriched" }, "how_to_use": { "text": "This dataset is intended for building and improving crop disease detection systems for smallholder farming in Ghana and similar West African contexts. It covers four crops --tomatoes, pepper and maize -- with 22 disease and health classes in total, making it one of the more comprehensive Afrocentric crop disease image collections available.\n\nYou can use this dataset to train image classification models that identify specific diseases from leaf photos. With nearly 25,000 raw images captured from local farms in Ghana (October-December 2022), plus over 100,000 augmented images with a ready-made train/test split, the dataset is structured for direct use in standard image classification workflows. The raw images are also available separately if you prefer to apply your own augmentation or splitting strategy.\n\nThe dataset is particularly valuable because it captures disease symptoms as they actually appear on farms in Ghana -- subtle, at various stages, and under real field conditions. This makes models trained on this data more likely to perform well in practical agricultural advisory tools than models trained on laboratory images. Agricultural technology developers, extension services, and research institutions can use it to build mobile apps or decision-support tools that help farmers identify and respond to crop diseases early.\n\nResearchers can extend this work by combining it with other crop disease datasets to improve cross-regional generalization, or by adding whole-plant and field-level imagery to complement the current leaf-level focus. The dataset is licensed under CC BY 4.0 and is available on Kaggle (approximately 20 GB). A free Kaggle account is required for download.\n\nKnown limitations: The images are from specific farming regions in Ghana, so models trained exclusively on this data may not generalize well to crops grown under different conditions elsewhere. The dataset focuses on leaf-level symptoms and does not include whole-plant or field-level imagery.\n\nSource: https://www.kaggle.com/datasets/responsibleailab/crop-disease-ghana", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_13/images/placeholder_image.jpeg" }, { "id": "ui_21", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_21-imarika__translating_weather_information_into/", "aliases": [ "imarika__translating_weather_information_into", "innovate_africa_challenge_ai_for_climate" ], "title": "Imarika - Translating weather information into actionable advisory for farmers through AI in Kenya", "description": { "text": "IMARIKA by Strathmore University’s iLabAfrica is building low-cost automatic weather station networks to provide access to accurate, local weather information in rural Kenya. This granular data allows advisory services for small-holder farmers to shift from using generic information to location-specific guidance, delivered through farmer organizations and digital platforms. IMARIKA also works on translating weather data into actionable recommendations, addressing farmers’ difficulty in interpreting weather information.\n\nIMARIKA by Strathmore University’s iLabAfrica is also the winner of the \"Innovate Africa Challenge on Climate Action\",  - chosen for its potential to advance climate adaptation for small-holder farmer communities. Together with Smart Africa, FAIR Forward had supported seven startups through intensive technical and business support to develop their AI businesses focused on mitigating and adapting to climate change in four countries through the Innovate Africa Challenge format in its first iteration focusing on AI for Climate Action.\n\nIMARIKA was also part of a Mozilla innovation challenge supporting people and projects across East Africa who leverage Common Voice’s open-source voice data set to unlock social and economic opportunities. These grants help to advance the use of open-source voice data for products that support community participation and engagement.", "provenance": "curated" }, "kind": [ "usecase" ], "countries": [ { "name": "Kenya", "iso2": "KE" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 10", "SDG 5" ], "data_types": [ "Meterological", "Text", "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": null, "organizations": { "provided_by": { "text": "Strathmore University’s iLabAfrica", "links": [] }, "catalyzed_by": { "text": "Smart Africa, FAIR Forward - AI for All, GIZ , Intellecap, intel.liftoff, Climate Change AI; Mozilla Foundation", "links": [ { "name": "Smart Africa", "url": "https://smartafrica.org/" }, { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ, Gates Foundation", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Betsy Muriithi , Strathmore University’s iLabAfrica", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "GitHub: imarika-weather-pipeline", "url": "https://github.com/iLab-DSU/imarika-weather-pipeline" }, { "label": "GitHub: Agricultural-Recommendations-Chat", "url": "https://github.com/iLab-DSU/Agricultural-Recommendations-Chat" } ], "additional": [ { "label": "Chatbots Local Weather Reports And A Boon For Kenyas Smallholder Farmers (foundation.mozilla.org)", "url": "https://foundation.mozilla.org/blog/chatbots-local-weather-reports-and-a-boon-for-kenyas-smallholder-farmers/" }, { "label": "Awards (mozillafoundation.org)", "url": "https://www.mozillafoundation.org/en/what-we-fund/programs/common-voice-kiswahili-awards/awards/" }, { "label": "Innovate Africa Challenge (smartafrica.org)", "url": "https://smartafrica.org/innovate-africa-challenge/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Weather data pipeline for rural Kenya. Fetches readings from the Wireless Planet API (api.wirelessplanet.co.ke) every 3 hours. License: MIT. Data Type: Time-series weather readings. Processing: Apache Spark 3.3.0 structured streaming with data cleaning, mean-based imputation, and Z-score anomaly detection (threshold: 10.0). Storage: PostgreSQL 14 (raw table: weather_raw; processed table: weather_clean with device_id, date, temperature, wind, precipitation, anomaly flags). Throughput: ~40-50 records/second, <30 seconds end-to-end latency. Daily aggregation compresses ~1,000 raw readings into ~35 daily summaries. Requires Docker, 8GB+ RAM, and a weather API account.\n\nSource: https://github.com/iLab-DSU/imarika-weather-pipeline", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Two components: (1) Weather Pipeline -- Apache Spark streaming processor ingests weather API data via Kafka, cleans it, detects anomalies, and stores raw + processed data in PostgreSQL. Produces daily weather summaries per device. Stack: Docker Compose, Kafka, Spark 3.3.0, PostgreSQL 14, Python. (2) Agricultural AI Assistant -- LangGraph-based advisory chatbot for 6 East African crops (beans, cassava, finger millet, maize, sorghum, sweet potatoes). Uses QLoRA-fine-tuned Llama 3.2 3B Instruct model, a knowledge graph (30+ nodes, 95+ edges), Chroma vector database, and OpenWeather API integration. Supports English and Swahili. Runs locally via Ollama. Requires 4GB+ GPU VRAM for inference or CPU (slower).\n\nSource: https://github.com/iLab-DSU/imarika-weather-pipeline, https://github.com/iLab-DSU/Agricultural-Recommendations-Chat", "provenance": "auto-enriched" }, "how_to_use": { "text": "For anyone interested in translating weather information into actionable advisory for farming practices, Imarika's Agricultural AI Assistant which is an intelligent agricultural advisory system powered by LangGraph, Knowledge Graph, and Ollama models specialized in 6 East African crops with multilingual support (English/Swahili) is worth looking into. \nAnyone running several weather stations with a need to process the data may be interested in Imarika's real-time weather data processing pipeline built with Apache Spark, Kafka, and PostgreSQL. \n\nThis use case also included the development of a business model and funding model for open source AI as a stepping stone towards financially viable operations.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_21/images/placeholder_image.jpeg" }, { "id": "ui_22", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_22-empowering_women_across_india_with_audio/", "aliases": [ "digital_audio_content_creation_for_womens", "empowering_women_across_india_with_audio" ], "title": "Empowering Women across India with audio messages in their native languages on Health, Sustainable Agriculture and Education", "description": { "text": "Empowering Women Across India with Voice-based Knowledge in Their Native Languages: By using openly accessible text-to-speech models from the Indian Institute of Science (IISc) and MeitY's Digital India Bhashini Division, Audiopedia created health-related 400+ audio messages that can be shared with civil society organizations for awareness raising and capacity building", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 10", "SDG 5" ], "data_types": [ "Text", "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "CC-BY-SA 4.0", "spdx": "CC-BY-SA-4.0", "url": null }, "organizations": { "provided_by": { "text": "Audiopedia", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Audiopedia Foundation (https://www.audiopedia.foundation/), Marcel Heyne (contact@audiopedia.org)", "access_note": null, "links": { "dataset": [ { "label": "audiopedia.app.box.com: 4wtqy4idpnilf3b3abisuuu2vdi0oi1q", "url": "https://audiopedia.app.box.com/s/4wtqy4idpnilf3b3abisuuu2vdi0oi1q" } ], "usecase": [ { "label": "audiopedia.foundation: bharat", "url": "https://www.audiopedia.foundation/bharat" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "Audiopedia has created useful content on various topics including health, education, financial literacy. This data is available in text and audio format in multiple languages incl. low-resource languages.", "provenance": "curated" }, "model_characteristics": { "text": "For the generation of audio messages, Audiopedia used openly accessible text-to-speech models from the Indian Institute of Science (IISc) and MeitY's Digital India Bhashini Division.", "provenance": "curated" }, "how_to_use": { "text": "Civil society organizations can use the existing voice content to generate tailored, AI-powered messages in various Indian languages across different topics including health, education, agriculture and other social topics. Civil society organizations can deploy these messages in rural and semi-urban areas using easy-to-use digital tools incl. WhatsApp messaging or feature phones.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_22/images/placeholder_image.jpeg" }, { "id": "ui_23", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_23-combatting_air_pollution_and_ghg_emissions/", "aliases": [ "combatting_air_pollution_and_ghg_emissions", "hyperlocal_mapping_of_air_pollution_and", "open_air_pollution_data_for_patna" ], "title": "Combatting Air Pollution and GHG Emissions in India through hyperlocal AI-powered mapping", "description": { "text": "Under this initiative, a novel approach is employed by leveraging citizen scientists and IoT-based low-cost sensors to collect hyperlocal air quality data. This data is used to identify pollution sources and risk zones, facilitating targeted actions by regulatory authorities.To showcase data outreach, the project features the VAYU Android-based application and the VAYU citizen portal digital stack, which support targeted interventions and customized solutions backed by AI/ML algorithms. These tools potentially develop new approaches in air pollution management while reducing public investment costs.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 13" ], "data_types": [ "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "UNDP India", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "UNDP India (registry.in@undp.org)", "access_note": null, "links": { "dataset": [ { "label": "vayu.undp.org.in", "url": "https://vayu.undp.org.in/" } ], "usecase": [ { "label": "GitHub: VAYU_OpenAir", "url": "https://github.com/undpindia/VAYU_OpenAir" }, { "label": "GitHub: vayu-gnn", "url": "https://github.com/EconAIorg/vayu-gnn/tree/main" }, { "label": "GitHub: VayuAssist", "url": "https://github.com/Alphawarrior21/VayuAssist" }, { "label": "GitHub: ClearSky", "url": "https://github.com/akbp24/ClearSky" }, { "label": "GitHub: vayu_airnode", "url": "https://github.com/sherwaldeepesh/vayu_airnode" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "The data for this project is collected by static and dynamic sensor. Static sensors are placed at known hotspots for air pollution while dynamic sensors are moved by citizen scientist continously. \n• 2 Cities\n• 100+ Senors\n• 150+ Volunteers\n• 1000+ Records Collected\n• ~10Million Data Points", "provenance": "curated" }, "model_characteristics": { "text": "Sensors data is continously shared with specific pollutoin boards, and us ecase dvelopeed under project combined AI and ML to detect pollution sources and provide early alerts, enabling rapid response and policy action.", "provenance": "curated" }, "how_to_use": { "text": "VAYU OpenAir provides an end-to-end open-source platform for hyperlocal air pollution mapping, built through a collaboration between UNDP, GIZ, the Government of India, the University of Nottingham, Development Alternatives, D-Coop, and citizen scientists across India. The platform includes open data, open algorithms, and open software from hyperlocal mapping campaigns in Patna and Gurgaon.\n\nThe core platform consists of three components: a mobile app for field-level data collection and viewing air quality readings, a web portal dashboard for visualising pollution data, and a backend API serving data to both interfaces. All source code is available at github.com/undpindia/VAYU_OpenAir under an MIT license.\n\nPractitioners can use VAYU OpenAir in several ways. If you are working on urban air quality in Indian cities or comparable contexts, you can deploy the existing platform to run your own hyperlocal mapping campaigns -- the mobile app supports citizen-science data collection, while the web portal provides ready-made visualisation tools for stakeholders and policymakers. If you already have air quality data and want to build predictive or analytical tools on top of it, several community-built open-source projects demonstrate what is possible:\n\n- **vayu-gnn** -- A Graph Neural Network that predicts hyperlocal pollutant levels up to 8 hours ahead by combining VAYU sensor readings with weather data, elevation, river distance, and urban density features. Available under GPL-3.0 at github.com/EconAIorg/vayu-gnn.\n- **VayuAssist** -- A RAG-based chatbot designed for government policymakers, providing air quality insights, AQI trend analysis, and mitigation strategies. Built with Streamlit and OpenAI GPT-3.5 Turbo. Available under MIT at github.com/Alphawarrior21/VayuAssist.\n- **ClearSky** -- Jupyter Notebooks for predictive pollution analysis with 3D visualisation. Available at github.com/akbp24/ClearSky.\n- **vayu_airnode** -- ARIMA time-series analysis for individual pollutants (CH4, CO, CO2, NO2, PM10, PM2.5) plus temperature and humidity, with a Streamlit interface. Available under Apache-2.0 at github.com/sherwaldeepesh/vayu_airnode.\n\nThese community tools illustrate concrete extension points: from short-term pollution forecasting to policy-support chatbots to time-series analysis of specific pollutants. Developers looking to extend the ecosystem can build on any of these as starting points.\n\nCost and resources: The platform and all community tools are open-source, so the primary costs are compute infrastructure for hosting the backend and any ML model training. Setup documentation is provided in each repository.\n\nSources:\n- https://vayu.undp.org.in/\n- https://github.com/undpindia/VAYU_OpenAir\n- https://github.com/EconAIorg/vayu-gnn/tree/main\n- https://github.com/Alphawarrior21/VayuAssist\n- https://github.com/akbp24/ClearSky\n- https://github.com/sherwaldeepesh/vayu_airnode", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_23/images/airpollution.png" }, { "id": "ui_24", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_24-making_ai_speak_9_indian_languages/", "aliases": [ "iisc__open_voice_data_in", "making_ai_speak_9_indian_languages" ], "title": "Making AI speak 9 Indian languages: Hindi, Bengali, Marathi, Telugu, Bhojpuri, Kannada, Magadhi, Chhattisgarhi, Maithili - Open-source text-to-speech models", "description": { "text": "This dataset is part of the initiative SYSPIN (SYnthesizing SPeech in INdian languages), that develops large open-source Text-to-Speech (TTS) corpora, i.e., speech training datasets and AI models in 9 Indian languages, namely – Bengali, Bhojpuri, Chhattisgarhi, Hindi, Kannada, Magahi, Maithili, Marathi, and Telugu. This data is useful for those wishing to build voice-enabled, user-friendly applications in Indian languages. Through voice-based solutions, these AI models can be used to empower communities that do not have access to digital services in their own native language. They can be used by private companies, researchers, NGOs or government departments to develop digital applications. \n \n The datasets and TTS models are available to the AI community and anyone wishing to build voice-enabled applications in Indian languages. Given the distribution of speakers. some of the languages are especially useful for programmes wishing to engage with rural communities and promote sustainable agriculture in rural settings.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 10" ], "data_types": [ "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Indian Institute of Science (IISc)", "links": [] }, "catalyzed_by": { "text": "Bhashini, FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Dr. Prasanta Ghosh (prasantag@iisc.ac.in)", "access_note": null, "links": { "dataset": [ { "label": "syspin.iisc.ac.in", "url": "https://syspin.iisc.ac.in/datasets" } ], "usecase": [ { "label": "Hugging Face: SYSPIN", "url": "https://huggingface.co/SYSPIN" }, { "label": "bhashini.gov.in: explore-models", "url": "https://bhashini.gov.in/ulca/model/explore-models" } ], "additional": [ { "label": "Loud And Clear Open Source For Indias Linguistic Diversity (bmz-digital.global)", "url": "https://www.bmz-digital.global/en/news/loud-and-clear-open-source-for-indias-linguistic-diversity/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The Data is fully open-sourced and can be accessed at:\n • IISc website: \n TTS Dataset:\n • License: CC-BY 4.0\n • Data Type: Audio recording\n • Sentence creation: Variety of domains covered in the sentences/text and phonetically rich sentence selection\n • Accounts for dialect variability\n • Voice artist selection and balanced duration per voice artist \n • Voice recording: 40 hours of recording by 1 male voice artist and 1 female voice artist for each of the 9 languages\n • Studio-quality audio with 48kHz, 24 bits per sample from every voice artist\n • Data was collected and created over 3 years from 2021-2024", "provenance": "curated" }, "model_characteristics": { "text": "Processed Data:\n The validated sentences were recording in a recording room (size 10'3\" x 5'9\") by voice artists using a Neumann TLM-103 studio microphone and Audio Interface UAD Apollo Twin X. \n Application:\n IISc has organised the LIMMITS challenges (2023, 2025) as part of IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) to build TTS voices, share web API for evaluation, and allow researchers to contribute towards the development of streaming and neural codec-based TTS systems. \n Audio files of Audiopedia Foundation’s content created in the low-resource language Chhattisgarhi using IISc TTS Chhattisgarhi model can be found here.", "provenance": "curated" }, "how_to_use": { "text": "The TTS models can be used to develop innovative solutions and voice-based services for Indians in their native language. This would be especially beneficial for those who cannot read/write or have speech and visual disabilities but can access digital services through audio/voice-based mediums. The data and models are a crucial stepping stone for developing such assistive technologies. Including low-resource languages, some of which do not even have enough print/digital literature, would benefit vulnerable communities for whom language barriers make access to technological solutions even more difficult. Given the distribution of speakers. some of the languages are especially useful for programmes wishing to engage with rural communities and promote sustainable agriculture in rural settings. The 720 hours of open-source TTS data also present immense opportunities for academic and industrial research.\n\nYou can directly access and use the models on Bashini here via web and API: https://anuvaad.bhashini.gov.in/", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_24/images/pexels-photo-18636912.jpg" }, { "id": "ui_25", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_25-open_source_ai_pest_control_for/", "aliases": [ "open_source_ai_pest_control_for", "wadhwani_ai__open_ai_for" ], "title": "Open source AI Pest Control for smallholder Cotton Farmers in India", "description": { "text": "Wadhwani AI has developed a mobile app to support cotton farmers combat pest infestations, a major threat to cotton productivity. For this, they have collaborated with FAIR Forward – AI for All, a project of German Development Cooperation. The app allows farmers to take photos of pheromone traps, which capture male bollworms, and uses an AI model to count the pests in the images. Based on the AI-generated counts, the app provides recommendations to farmers on how to respond. This technology empowers small-scale farmers to make informed decisions, reduce uncertainty in yield, and improve their livelihoods.\n \n As the codebase, dataset and research are openly available it can be used to build on, test and research similar applications in other contexts. This would esp. apply to problems of identification and counting of agricultural pests.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 2" ], "data_types": [ "Images" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": { "name": "Apache 2.0", "spdx": "Apache-2.0", "url": null }, "organizations": { "provided_by": { "text": "Wadhwani AI", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Wadhwani AI (agri-ai@wadhwaniai.org), Github (https://github.com/WadhwaniAI/pest-monitoring)", "access_note": null, "links": { "dataset": [ { "label": "source.coop: description", "url": "https://source.coop/repositories/wadhwani-ai/wiai-pm-open-data/description" }, { "label": "GitHub: pest-management-opendata", "url": "https://github.com/WadhwaniAI/pest-management-opendata?tab=readme-ov-file" }, { "label": "Hugging Face: pest-management-opendata", "url": "https://huggingface.co/datasets/wadhwani-ai/pest-management-opendata" } ], "usecase": [ { "label": "GitHub: pest-monitoring", "url": "https://github.com/WadhwaniAI/pest-monitoring" } ], "additional": [ { "label": "Pest Management Ai Solution (wadhwaniai.org)", "url": "https://www.wadhwaniai.org/programs/pest-management/pest-management-ai-solution/" }, { "label": "3394486.3403363 (dl.acm.org)", "url": "https://dl.acm.org/doi/10.1145/3394486.3403363" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Wadhwani AI with support from German Development Cooperation provided training data consisting of ca. 13,000 images:\n \n (a) These images were captured by farmers and farm extension workers since 2018\n (b) The dataset contains two types of images: those with pests and bounding boxes labelling either pink bollworm (PBW) or American bollworm (ABW), and those without pests, representing real-world user behavior.\n (c) License: Apache 2.0", "provenance": "curated" }, "model_characteristics": { "text": "Wadhwani AI with support from German Development cooperation provided the codebase of the pest identification model:\n \n (a) At its core, this repository packages an object detection implementation. \n (b) While there are several object detection implementations, and even implementation aggregations, there are none that completely solve for the challenges faced here (e.g., highly diverse image quality, model size restrictions)\n (c) License: Apache 2.0", "provenance": "curated" }, "how_to_use": { "text": "The resources can be used to build on, test and research similar applications in other contexts, esp. applying to problems of identification and counting of agricultural pests.\n\nThis use case also included the development of a business model and funding model for open source AI as a stepping stone towards financially viable operations. It was facilitated by Villgro Africa's and FAIR Forward's six-month mentorship programme on \"Creating sustainable business and funding models with open source AI\". See: https://www.bmz-digital.global/en/news/open-source-ai-business-impact/", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_25/images/placeholder_image.jpeg" }, { "id": "ui_26", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_26-providing_better_information_on_sexual_and/", "aliases": [ "gramvaani_automatic_speech_recognition_asr", "providing_better_information_on_sexual_and" ], "title": "Providing better information on sexual and reproductive health and rights of young people through the Kahi Ankahi Baatein infoline (Hindi LLM finetuning)", "description": { "text": "Improve the Kahi Ankahi Baatein (KAB) platform by fine-tuning Hindi LLMs for better user experience.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 10" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": null, "organizations": { "provided_by": { "text": "GramVaani", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Gram Vaani (contact@gramvaani.org)", "access_note": null, "links": { "dataset": [ { "label": "GitLab: SRHR automated QnA-KAB.xlsx", "url": "https://gitlab.com/gramvaani/giz_kab_bert_sourcecode/-/blob/main/SRHR%20automated%20QnA-KAB.xlsx?ref_type=heads" } ], "usecase": [ { "label": "GitLab: giz_kab_bert_sourcecode", "url": "https://gitlab.com/gramvaani/giz_kab_bert_sourcecode" } ], "additional": [ { "label": "Answering Questions On Sexuality Through Kahi Ankahi Baatein 2 (gramvaani.org)", "url": "https://gramvaani.org/answering-questions-on-sexuality-through-kahi-ankahi-baatein-2/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The dataset provides a small sample of manually transcribed voice data from the KAB helpline (https://gramvaani.org/indias-new-sexual-health-info-line/). The voice recordings are in Hindi as well as the transcriptions. The dataset consists of transcriptions of Questions that people have asked as well as the corresponding answers from the database. The dataset is further tagged with other useful labels such as domain or relevance of answer provided.", "provenance": "curated" }, "model_characteristics": { "text": "A Transformer-based architecture, specifically BERT (bert-base-multilingual-cased), is employed using TensorFlow. The model is trained on question pairs for a similarity classification task. Positive samples consist of similar question pairs, while negative samples involve random question pairs. The pretrained multilingual BERT model from HuggingFace is utilized to obtain embeddings for each input.Eventually, as a result the model generates similarity scores for question pairs in the test set.", "provenance": "curated" }, "how_to_use": { "text": "There are two resources to build on and use:\n1) The dataset can be used as a resource to fine-tune Q&A capabilities of LLMs specifically for the Sexual and Reproductive Health domain. The dataset can be further used to quality check and/or improve Speech to Text models. \n2) the fine-tuned model can be used for down-stream applications that need Hindi LLM capabilities in the Sexual and Reproductive Health domain.\n\nThis use case also included the development of a business model and funding model for open source AI as a stepping stone towards financially viable operations. It was facilitated by Villgro Africa's and FAIR Forward's six-month mentorship programme on \"Creating sustainable business and funding models with open source AI\". See: https://www.bmz-digital.global/en/news/open-source-ai-business-impact/", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_26/images/placeholder_image.jpeg" }, { "id": "ui_27", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_27-making_ai_speak_mundari__opensource/", "aliases": [ "100_hours_of_text_to_speech", "making_ai_speak_mundari__opensource" ], "title": "Making AI speak Mundari - Open-source text-to-speech data and model for Mundari, India", "description": { "text": "100 hours of Text to Speech Dataset for Mundari Language", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 10" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model", "tags": [ "dataset", "model" ] }, "license": { "name": "BY-NC-SA-FS", "spdx": null, "url": null }, "organizations": { "provided_by": { "text": "Karya Inc.", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Karya (operations@karya.in)", "access_note": null, "links": { "dataset": [ { "label": "GitHub: dataset-hindi-mundari-translation", "url": "https://github.com/karya-inc/dataset-hindi-mundari-translation" }, { "label": "GitHub: dataset-mundari-tts", "url": "https://github.com/karya-inc/dataset-mundari-tts" } ], "usecase": [ { "label": "GitHub: MunTTS-A-Text-to-Speech-System-Fo…", "url": "https://github.com/microsoft/MunTTS-A-Text-to-Speech-System-For-Mundari" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "Translation: This corpus contains a collection of sentence pairs that have been translated from Hindi to Mundari. The dataset was created as part of a collaboration between Microsoft Research India, Indian Institute of Technology Kharagpur, and Karya. This dataset is realeased under the non-commercial version of the Karya Public License. Please read the \"License\" section below for a high-level summary of what you are allowed and not allowed to do under this license. The dataset contains 17,826 sentence pairs of Hindi to Mundari translations. The translations are stored in a single TSV file translation-hi-unr.tsv with UTF-8 encoding. Each line in the TSV file contains two columns, separated by a tab character. The first column contains the Hindi sentence, and the second column contains the corresponding Mundari translation.\n\nText-to-speech:The Mundari TTS dataset corpus contains a total of 26,870 audio files, each containing a single utterance spoken by one of the two speakers. The audio is recorded in 32-bit PCM format with a sampling rate of 44.1 kHz. The dataset includes transcripts for each audio file in Mundari script. The recordings were collected in a sound-treated room using a high-quality microphone and preamp.", "provenance": "curated" }, "model_characteristics": { "text": "MunTTS, an end-to-end text-to-speech (TTS) system specifically for Mundari, a low-resource Indian language of the Austo-Asiatic family. Our work addresses the gap in linguistic technology for underrepresented languages by collecting and processing data to build a speech synthesis system. We begin our study by gathering a substantial dataset of Mundari text and speech and train end-to-end speech models. We also delve into the methods used for training our models, ensuring they are efficient and effective despite the data constraints. We evaluate our system with native speakers and objective metrics, demonstrating its potential as a tool for preserving and promoting the Mundari language in the digital age.", "provenance": "curated" }, "how_to_use": { "text": "The Mundari TTS dataset corpus includes recordings from two different speakers: a female speaker and a male speaker. The female speaker contributed 19,868 recordings, while the male speaker contributed 7,002 recordings.\r\n\r\nThe total size of the corpus is approximately 17 GB (7 GB compressed). Therefore, the full dataset is hosted in cloud storage (instead of this github repository). This repository contains a sample of 100 recordings from each speaker. Please email data@karya.in for a link to download the full dataset.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_27/images/placeholder_image.jpeg" }, { "id": "ui_28", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_28-farmerchat_delivering_personalized_farm_advice_to/", "aliases": [ "farmerchat_delivering_personalized_farm_advice_to", "farmerchat_open_aipowered_agricultural_advisory_fo" ], "title": "FarmerChat: Delivering Personalized Farm Advice to 1.6 Million Farmers Across Five Countries", "description": { "text": "FarmerChat is an AI assistant built to help smallholder farmers make better field-level decisions by delivering timely, localized advice to help them grow more, earn more, and adapt to changing weather. It is designed for low-connectivity areas and available in over 15 languages including Swahili, Portuguese, Spanish, Hindi, Telugu, Amharic, Hausa, and English.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" }, { "name": "Kenya", "iso2": "KE" }, { "name": "Nigeria", "iso2": "NG" }, { "name": "Ethiopia", "iso2": "ET" }, { "name": "Brazil", "iso2": "BR" } ], "regions": [], "sdgs": [ "SDG 2" ], "data_types": [ "Other" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Digital Green", "links": [] }, "catalyzed_by": { "text": "Digital Green, in parts by FAIR Forward - AI for All, GIZ", "links": [ { "name": "in parts by FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "Multiple Funders, among them: The Rockefeller Foundation, Gates Foundation, BMZ and Sequoia Climate Fund, CISCO Foundation", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Digital Green (contact@digitalgreen.org)", "access_note": null, "links": { "dataset": [ { "label": "GitHub: DG_Open", "url": "https://github.com/digitalgreenorg/DG_Open" }, { "label": "Hugging Face: farmerchat-queries-large", "url": "https://huggingface.co/datasets/DigiGreen/farmerchat-queries-large" } ], "usecase": [ { "label": "play.google.com: details", "url": "https://play.google.com/store/apps/details?id=org.digitalgreen.farmer.chat&hl=en_US" } ], "additional": [ { "label": "View (drive.google.com)", "url": "https://drive.google.com/file/d/11jmgluDCSqyMzP9s_VJCMgZVGkdmXePH/view?usp=sharing" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The FarmerChat system draws from an agronomic content corpus that powers FarmerChat's advisory responses — curated from ICAR, national extension services, and field-validated crop guidance across India, Kenya, Ethiopia, Nigeria, and Brazil. This content is organized by crop, geography, and query type, enabling the system to return locally grounded answers rather than generic advice.\n\nKey characteristics: multilingual (15 languages), multi-country (5 geographies), low-bandwidth optimized, and continuously updated through fine-tuning and RLHF loops informed by farmer feedback and agronomist review. Query volume has reached 9 million across 1.6 million farmers, providing a rare at-scale signal on what smallholder farmers actually ask and what advice they act on.\n\nResponsible AI: FarmerChat's content pipeline includes human agronomist review before deployment in any new crop or geography. Outputs are grounded in verified extension material to reduce hallucination risk. Digital Green has invested in RLHF infrastructure to surface low-quality responses for expert correction.", "provenance": "curated" }, "model_characteristics": { "text": "Multiple models are available based on the shared data. Please refer to the GitHub repository.", "provenance": "curated" }, "how_to_use": { "text": "FarmerChat can be engaged at four levels depending on your organization's technical capacity, timeline, and deployment goals.\n\nOpen Build Partner: Technically equipped organizations can build independently on Digital Green's public infrastructure. The full codebase is available at github.com/digitalgreenorg/DG_Open (Apache 2.0) and datasets are on GitHub and Hugging Face. Digital Green publishes and maintains the open source foundation; the partner owns everything from development through deployment.\n\nCommunity Partner: For organizations ready to launch quickly without requiring Digital Green staff time. Digital Green provides and maintains the FarmerChat platform at no cost, along with an onboarding toolkit, co-brandable campaign templates, and remote training support. The partner leads all in-country training and farmer outreach.\n\nCustom Partnership: For organizations that want Digital Green staff actively involved in their deployment. Depending on scope, this can range from research-grade validation — where Digital Green contributes expert agronomic content review, M&E frameworks, and pre-built analytics dashboards — through to full co-implementation, where Digital Green dedicates product and engineering staff to customize across various features of localization and the tech stack.\n\nTo get started, organizations should reach out to Digital Green to find the partnership tier that best suits their needs.\n\nBasic eligibility: farmers should be reachable in a language FarmerChat supports (including English, Hindi, Telugu, Amharic, Kiswahili, Hausa, Portuguese, Spanish, and others), have access to an Android smartphone directly or via a shared-device model, and the project should have a credible distribution path through an extension network, cooperative, government department, or telecom partnership.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_28/images/digital_green.png" }, { "id": "ui_29", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_29-predicting_crop_health_using_opensource_geospatial/", "aliases": [ "ml4eo_telangana_crop_identification", "open_dataset_of_farm_boundaries_and", "predicting_crop_health_using_opensource_geospatial" ], "title": "Predicting Crop Health using open-source geospatial data and ground truth data collected in Telangana State", "description": { "text": "This AI application and replication-kit is about an AI-based crop type map for Telangana. This should be of use to anyone wishing to support sustainable farming practices and evidence-based decision-making for agricultural. Also, this can be used to empower you, as a local business, and NGOs, or a government agency to develop digital agricultural applications. The application focuses on creating localized ground-truth data on several different crop related variables to train AI models tailored to Telangana's diverse agricultural landscape. This includes an open-access ground-truth dataset, a data challenge with openly available models for yield prediction, a crop classification model using satellite imagery applied to map the whole state of Telangana, and a roadmap for AI implementation in agriculture. Datasets, AI models, and crop maps are all openly available as an open-source replication kit and digital public good.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 2" ], "data_types": [ "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": null, "organizations": { "provided_by": { "text": "WRMS - Weather Risk Management Services, ZINDI, Vertify Earth", "links": [ { "name": "WRMS - Weather Risk Management Services", "url": "https://wrmsglobal.com/" }, { "name": "ZINDI", "url": "https://zindi.africa/" }, { "name": "Vertify Earth", "url": "https://vertify.earth/" } ] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "WRMS (connect@wrmsglobal.com)", "access_note": null, "links": { "dataset": [ { "label": "dataexplorer.ts.adex.org.in: 583e8f01-160e-4f51-bde5-31dc7f2a5…", "url": "https://dataexplorer.ts.adex.org.in/dataset/583e8f01-160e-4f51-bde5-31dc7f2a5887" } ], "usecase": [ { "label": "GitHub: zindi_crop_health", "url": "https://github.com/pranavmyname/zindi_crop_health" }, { "label": "zindi.africa: telangana-crop-health-challenge", "url": "https://zindi.africa/competitions/telangana-crop-health-challenge" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "The Data is fully open-sourced and can be accessed at the Telangana Data Exchange Platform: ADEX\n • License: CC BY 4.0\n • Data Type: Tabular\n • Key Variables: Crop type (6), Crop Health, Crop Yield, Irrigation Methods.\n • Appr. No of observations: > 12000 \n • Appr. No of dimensions: 45 \n • Covering 6 Mandals in Telangana\n • Format: dbf, csv\n \n The data was collected over the span of 1.5 years covering different crop cycles between 2022 and 2024. The Data Collection was conducted by WRMS via on-the-ground visits of the fields and by surveying farmers in Telangana State.\n\nA responsible AI Assessment was undertaken for this dataset / use case to help AI developers and project managers to identify, assess and mitigate potential harms and biases in AI. For methodology, see https://www.bmz-digital.global/en/news/ethical-crash-test-for-ai-how-to-navigate-the-road-to-responsible-innovation/", "provenance": "curated" }, "model_characteristics": { "text": "Processed Data:\n The ground truth data was combined with open-source Sentinel-2 data to create a geospatial dataset with crop type as labels.\n The processed dataset can be accessed here: Processed Data\n Code for Crop Classification Model:\n Code to create the training data and for training the models for crop mapping can be accessed here: Code\n Application:\n The resulting shapefiles that predict crop type for the state of Telangana are deployed on DICRA – a digital public good ran by the Indian Development Bank NABARD and UNDP: DICRA\n Based on the collected data, GIZ organised a data challenged together with Zindi to build state-of-the-art crop mapping models. The outcomes of the challenge can be accessed here: Zindi", "provenance": "curated" }, "how_to_use": { "text": "You can make use of the existing models on yield prediction and crop type mapping to infer other similar datapoints – for example neighbouring states of Telangana. Additionally, you can try to combine the dataset with other existing open-source datasets from the ADEX platform and conduct research/commercial scoping on these. \n The present dataset is quite unique in the fact that it includes observations such as irrigation methods, crop health and expected yield. You may combine these with other data sources, i.e. satellite/remote sensing data (open-source or commercial) to build high-quality crop monitoring/recommendation systems to support various agricultural entities, farmers and decision makers in India. This could also inform commercial products. \n As crop mapping is a prominent use-case, the data can be used for benchmarking different available crop datasets and models or inform foundational geospatial models as well as constitute to a region-wide crop mapping. Potential partners on the ground for collaboration, commercial involvement or funding are the Telangana Government, specifically the IT and Agricultural department and Nabard Dicra.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_29/images/placeholder_image.jpeg" }, { "id": "ui_31", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_31-forest_forward_ii_using_ai_to/", "aliases": [ "ai_for_forest_conservation", "forest_forward", "forest_forward_ii_using_ai_to" ], "title": "Forest Forward II: Using AI to map carbon in forests using High Carbon Stock Approach and assess fire vulnerability of forests to Combat Climate Change and Protect Livelihoods in India", "description": { "text": "Mapping carbon content in forests: This application leverages advanced geospatial technologies, such as remote sensing and AI, to support forest conservation efforts in India. The application will allow stakeholders to better monitor forest health, estimate biomass, and contribute to climate mitigation efforts.\n Fire Vulnerability Mapping: This application would help develop a fire vulnerability map which could help to identify areas that are prone to fires based on several forest degradation indicators.", "provenance": "curated" }, "kind": [ "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 13" ], "data_types": [ "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "MIT", "spdx": "MIT", "url": null }, "organizations": { "provided_by": { "text": "Vertify Earth", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Michael Anthony ", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "GitHub: biomass-dl-model-training", "url": "https://github.com/vertify-earth/biomass-dl-model-training" }, { "label": "GitHub: ForestFireGoa", "url": "https://github.com/vertify-earth/ForestFireGoa" }, { "label": "Hugging Face: biomass-model", "url": "https://huggingface.co/vertify/biomass-model" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "The datasets used for this projects are all publicly available or accessible via common platforms such as Planet.com \nMore details to follow.", "provenance": "curated" }, "model_characteristics": { "text": "This model is trained to predict above-ground biomass (AGB) in tropical and subtropical forests using multi-source satellite imagery at the pixel level. Specifically:\r\n\r\nPrediction Unit: Estimates biomass at individual pixel level (typically 10-40m resolution depending on input data)\r\nOutput: Biomass density in Mg/ha (megagrams per hectare)\r\nInput Data: Processes multi-sensor data including Sentinel-1, Sentinel-2, Landsat-8, PALSAR, and DEM\r\nApplication Scope: Best suited for tropical and subtropical forest ecosystems in South/Southeast Asia\r\nBiomass Range: Validated for forests with biomass between ~40-460 Mg/ha", "provenance": "curated" }, "how_to_use": { "text": "The pipeline is designed to handle multi-source satellite imagery and corresponding biomass data with pixel-level precision. Key aspects include:\r\n\r\nEnd-to-End Workflow: From raw data ingestion to model training, evaluation, and deployment\r\nAdvanced Feature Engineering: Comprehensive spectral indices, texture features, spatial gradients, and PCA components\r\nStable Neural Architecture: Utilizes a custom StableResNet with residual connections and layer normalization for robust biomass regression\r\nMulti-Site Data Processing: Capable of processing and integrating data from multiple geographically distinct study sites\r\nFlexible Deployment: Includes HuggingFace deployment capabilities with Gradio interface\r\nMemory Efficient Processing: Chunk-based processing for handling large satellite images", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_31/images/goa-forest.jpg" }, { "id": "ui_33", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_33-voice_tech_for_all_building_inclusive/", "aliases": [ "making_the_indian_voice_datasets_more", "voice_tech_for_all_building_inclusive" ], "title": "Voice Tech for All: Building Inclusive Speech AI for India with different accents and speaking styles", "description": { "text": "This hackathon challenged teams to build an AI system that can turn written text into natural-sounding speech across multiple Indian languages, including less-represented ones. Participants created voices that can reflect different accents and speaking styles, making the technology more personal and inclusive. Using open data, they made voice technology more accessible for diverse communities. The solutions will be integrated into a voice assistant designed to support pregnant women in low-income communities.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 10" ], "data_types": [ "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Indian Institute of Science (IISc), Artpark", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Dr. Prasanta Ghosh (prasantag@iisc.ac.in)", "access_note": null, "links": { "dataset": [ { "label": "spiredatasets.ee.iisc.ac.in: syspincorpus", "url": "https://spiredatasets.ee.iisc.ac.in/syspincorpus" }, { "label": "spiredatasets.ee.iisc.ac.in: englishttscorpus", "url": "https://spiredatasets.ee.iisc.ac.in/englishttscorpus" }, { "label": "spiredatasets.ee.iisc.ac.in: gujaratittscorpus", "url": "https://spiredatasets.ee.iisc.ac.in/gujaratittscorpus" } ], "usecase": [ { "label": "Hugging Face: Spark_somya_TTS", "url": "https://huggingface.co/somyalab/Spark_somya_TTS" }, { "label": "Hugging Face: Voice-Tech-CDAC-Submission", "url": "https://huggingface.co/roymukund/Voice-Tech-CDAC-Submission/tree/main" }, { "label": "Hugging Face: voice-tech-for-all-v2", "url": "https://huggingface.co/vinaybabu/voice-tech-for-all-v2" }, { "label": "Hugging Face: voice-tech-for-all-challenge-v2", "url": "https://huggingface.co/immverse-ai/voice-tech-for-all-challenge-v2" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "\"The Data is fully open-sourced and can be accessed at:\n • IISc website: \n TTS Dataset:\n • License: CC-BY 4.0\n • Data Type: Audio recording\n • Sentence creation: Variety of domains covered in the sentences/text and phonetically rich sentence selection\n • Accounts for dialect variability\n • Voice artist selection and balanced duration per voice artist \n • Voice recording: 40 hours of recording by 1 male voice artist and 1 female voice artist for each of the 9 languages\n • Studio-quality audio with 48kHz, 24 bits per sample from every voice artist\n • Data was collected and created over 3 years from 2021-2024\"", "provenance": "curated" }, "model_characteristics": { "text": "The winning models can be accessed on their respective repositories here: \nSomya: https://huggingface.co/somyalab/Spark_somya_TTS\n\nCDAC_SVNIT: https://huggingface.co/roymukund/Voice-Tech-CDAC-Submission/tree/main\n\nLTRC-SPL: https://huggingface.co/vinaybabu/voice-tech-for-all-v2\n\nimmverse_ai: https://huggingface.co/immverse-ai/voice-tech-for-all-challenge-v2", "provenance": "curated" }, "how_to_use": { "text": "The TTS models can be used to develop innovative solutions and voice-based services for Indians in their native language. This would be especially beneficial for those who cannot read/write or have speech and visual disabilities but can access digital services through audio/voice-based mediums. The data and models are a crucial stepping stone for developing such assistive technologies. Including low-resource languages, some of which do not even have enough print/digital literature, would benefit vulnerable communities for whom language barriers make access to technological solutions even more difficult. Given the distribution of speakers. some of the languages are especially useful for programmes wishing to engage with rural communities and promote sustainable agriculture in rural settings.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_33/images/placeholder_image.jpeg" }, { "id": "ui_34", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_34-aipowered_monitoring_of_forest_degradation_and/", "aliases": [ "aipowered_monitoring_of_forest_degradation_and" ], "title": "AI-powered monitoring of forest degradation and impact of restoration programs in India's Eastern Himalayas", "description": { "text": "An AI-Driven Dataset for Nature-Positive Livelihoods and Forest Restoration in Eastern Himalayas.This open-access dataset and digital MRV (Monitoring, Reporting, and Verification) toolkit was developed to help communities, NGOs and policymakers in the Eastern Himalayas measure and manage the links between forests, livelihoods and climate resilience. The project was created in response to a persistent data gap in the Himalayan region where limited open data and difficult terrain have hindered effective monitoring of forest degradation and the impact of restoration programs. Using locally calibrated AI models trained on Indian biomass data, the system combines satellite observations with ground level socio-economic surveys to measure forest dependency, identify degradation hotspots and track restoration outcomes. All datasets ranging from community level livelihood surveys in Sikkim and Mizoram to forest biomass and land use layers in Assam are openly available. Together they form a replicable, open-source “digital public good” that can be used to train local AI models for forest monitoring, biodiversity accounting or impact assessment. By providing a scalable, context-aware dataset, this initiative enables indigenous start-ups, environmental NGOs and impact investors to quantify nature positive outcomes, align with IRIS+ indicators, and guide conservation investments.\nThe goal is to make environmental monitoring inclusive and data driven, empowering local actors to track changes in forest carbon, identify erosion risks and measure the socio-economic benefits of restoration. Ultimately, this approach helps governments, researchers and entrepreneurs build locally adapted AI systems that support both ecosystem regeneration and sustainable livelihoods in fragile mountain regions.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 15", "SDG 13" ], "data_types": [ "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Vertify Earth, Earth Analytics India Pvt Ltd, Balipara Foundation, Alsisar Impact", "links": [] }, "catalyzed_by": { "text": "Lacuna-Fund / Meridian (Climate-call) & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Vertify.earth - Michael Anthony (michael@vertify.earth), \nAlsisar Impact (anuj@alsisarimpact.com )", "access_note": null, "links": { "dataset": [ { "label": "GitHub: biomass-prediction-NorthEastIndia", "url": "https://github.com/vertify-earth/biomass-prediction-NorthEastIndia" }, { "label": "Zenodo: 16536024", "url": "https://zenodo.org/records/16536024" } ], "usecase": [ { "label": "felt.com: Nature-Audit-on-Agapi-Sourcing-Si…", "url": "https://felt.com/embed/map/Nature-Audit-on-Agapi-Sourcing-Sites-JBeP3LpnTCmabCOFb9AVrAC?loc=27.317897%2C88.595673%2C15.25z" } ], "additional": [ { "label": "Edit (docs.google.com)", "url": "https://docs.google.com/spreadsheets/d/1WaunaB-HU-j9UDs-X5Fn1vJ15H4ENMe-lcEgWM9l3Eg/edit?usp=sharing" } ], "documents": [] }, "content": { "data_characteristics": { "text": "This dataset integrates geospatial and socio-economic information to monitor forest restoration and livelihoods in the Eastern Himalayas. It combines satellite data from Sentinel-1, Sentinel-2, Landsat-8, and PALSAR with 500+ ground truth points and nearly 300 household surveys collected in Assam, Sikkim, and Mizoram. The data support AI applications for biomass estimation, deforestation detection and nature positive impact assessment.\nA locally trained biomass model improves accuracy for South Asian forest types, addressing calibration bias in global datasets. All personal identifiers have been removed, and sensitive coordinates anonymized to protect community privacy. The dataset is published under a CC-BY 4.0 license and maintained by Vertify.earth GmbH and Earth Analytics India Pvt Ltd, ensuring its ongoing usability as an open and replicable foundation for environmental AI and digital MRV systems.", "provenance": "curated" }, "model_characteristics": { "text": "The AI model estimates forest biomass and vegetation health in the Eastern Himalayas using multi-source satellite data combined with local ground measurements. It outputs Above-Ground Biomass (AGB) maps, enabling accurate tracking of forest carbon, degradation, and restoration progress.\nThe model accepts pre-processed satellite imagery as input and produces spatial biomass layers compatible with MRV dashboards. It runs on standard Python environments and can be adapted for other tropical regions with local calibration.\n\nCalibrated for South Asian forests; retraining is needed for other regions. Seasonal cloud cover can affect results. All training data are anonymized, and community consent was obtained. Users should validate outputs locally before application. Released under CC-BY 4.0 for open reuse and replication. Please credit Vertify.earth and Earth Analytics India and share improvements through the Vertify.earth GitHub", "provenance": "curated" }, "how_to_use": { "text": "This dataset and AI model can be used immediately for forest monitoring, restoration planning and impact measurement in mountain and forested regions. Users can map biomass, detect forest loss or assess community-level livelihood impacts by combining the open dataset with freely available satellite imagery. NGOs, research institutions, investors can directly apply the dataset to design or evaluate nature positive projects, while developers can integrate the AI model into their own MRV dashboards. Researchers and AI developers can extend this work by retraining the biomass model with local field data from other regions or adding new layers such as soil moisture or biodiversity indicators. Ethical replication should include community consent and data validation steps to ensure fair and context aware use.\n\nData access: Free (CC-BY 4.0) via Vertify.earth GitHub\r\n\r\nModel use: Free, standard cloud or local compute", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_34/images/placeholder_image.jpeg" }, { "id": "ui_35", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_35-combatting_climate_disinformation_in_indonesian_la/", "aliases": [ "combatting_climate_disinformation_in_indonesian_la", "use_case_development_indonesia" ], "title": "Combatting Climate Disinformation in Indonesian languages with AI", "description": { "text": "This AI application is about developing an AI-based system to tackle climate misinformation in Indonesia, focusing on creating accurate and accessible information tools. The initiative aims to build a multilingual disinformation detection system, covering Bahasa Indonesia and over 10,6 million words and 21 million words translation from Minangkabau, Balinese, and Buginese to classify and verify climate-related content. The application supports stakeholders, including government agencies, NGOs, and communities, in countering false narratives, promoting awareness, and enabling evidence-based decision-making. The AI outcomes include open-access datasets, disinformation, and topic classification models, a user-friendly website, and a chatbot. These tools empower users to identify climate misinformation, enhance public literacy, and foster collaboration for sustainable climate actions. The project emphasizes inclusivity, accessibility, and long-term impact, providing resources for researchers, local businesses, and policy-makers to advance AI applications in combating misinformation. Prosa has developed an AI-powered chatbot to address climate misinformation. The platform's website is available at faktaiklim.prosa.ai, and the chatbot can be accessed via Telegram (Fakta Iklim Bot), with open-source code available on GitHub. User guides are provided in four languages: Indonesian, Minangkabau, Balinese, and Bugis.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Indonesia", "iso2": "ID" } ], "regions": [], "sdgs": [ "SDG 10" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Prosa.ai", "links": [] }, "catalyzed_by": { "text": "Bappenas - Ministry of State Development Planning (Indonesia), FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Prosa AI (business@prosa.ai)", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: prosa-text", "url": "https://huggingface.co/prosa-text" } ], "usecase": [ { "label": "faktaiklim.prosa.ai", "url": "https://faktaiklim.prosa.ai/" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "The data for this project is fully open-sourced and can be accessed through Hugging Face.\n ● License: CC BY-SA 4.0\n ● Data Type: Tabular\n ● Key Variables: Climate disinformation classification, topic categorization, language distribution (Bahasa Indonesia and regional languages: Minangkabau, Balinese, Buginese)\n ● Format: CSV\n ● Data Collection Period: Over 3 months, covering diverse misinformation cases in climate-related domains\n \n Data collection involved web crawling and inputs from local communities and experts, using sources like government websites, news articles, and scientific literature. Annotations by trained linguists and marginalized community members followed strict guidelines for consistency. The datasets are formatted for easy integration with AI tools.", "provenance": "curated" }, "model_characteristics": { "text": "Processed Data and Pretrained Models:\n Use the multilingual datasets and pre-trained models (IndoBERT, IndoBERTNusa) to detect climate misinformation. The processed dataset can be accessed here: Processed Data\n \n Code for AI Models:\n Access the GitHub repository containing scripts for preprocessing datasets, fine-tuning language models, and training disinformation classifiers. These can be customized for various applications.\n \n Application Deployment:\n The resulting models are integrated into platforms like Kominfo's Anti-Hoax portal and social media crawling APIs. Additionally, a chatbot leveraging these AI models is available to verify climate-related information in real-time. The chatbot and models are publicly accessible via APIs to support further innovation and integration by developers.", "provenance": "curated" }, "how_to_use": { "text": "You can leverage AI models and datasets to develop or enhance applications to address climate misinformation and advance public education. These resources can be adapted to new languages, regions, or even other misinformation topics, such as health or political issues. Combining the datasets with additional sources, like satellite imagery or weather data, could lead to innovative AI applications, including real-time fact-checking tools or misinformation monitoring dashboards. Researchers and developers can also use the models to benchmark against other datasets or refine methodologies, unlocking opportunities for both academic and commercial advancements.\n \n These tools can be integrated into platforms like government portals or social media moderation systems to maximize impact. Interactive educational applications, such as gamified platforms or public learning modules, could further engage communities. Collaboration with stakeholders—including government agencies, NGOs, and private companies—can help scale the tools, secure funding, and drive wider adoption. Through partnerships and expanded deployment, this project can contribute to fostering critical awareness and combating misinformation effectively.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_35/images/placeholder_image.jpeg" }, { "id": "ui_36", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_36-high_carbon_stock_approach_mapping_forests/", "aliases": [ "hcsa_high_carbon_stock_mapping_in", "high_carbon_stock_approach_mapping_forests" ], "title": "High Carbon Stock Approach: Mapping Forests to Combat Climate Change and Protect Livelihoods in Indonesia", "description": { "text": "As the world's largest tropical rainforest, Indonesia’s forests are disappearing faster than decision-makers can respond - largely due to the lack of accessible and trusted data. Our project delivers validated, open-source High Carbon Stock forest maps, using AI and satellite imagery to identify high-carbon, high-biodiversity forests, regenerating forests, and low-carbon land suitable for sustainable development.\n\nThe data is public, transparent, and scalable, empowering NGOs, indigenous communities, and governments to prevent deforestation and reduce emissions into actionable classes:\n\nHigh, Medium, and Low-Density Forests (Conservation Priority):\nForests with closed or patchy canopies, high biodiversity value, and significant carbon storage that are essential for climate mitigation and ecosystem protection.\n\nYoung Regenerating Forests:\nTransitional forests with growing carbon sequestration potential, requiring forward-looking protection strategies to prevent future deforestation.\n\nScrub and Open Land:\nLow-carbon areas suitable for sustainable development, enabling economic growth without compromising forest integrity.\n\nImpact & Use Cases\n\nThe resulting maps empower:\n\n- Policymakers: to strengthen spatial planning and climate-aligned land-use policies\n\n- NGOs: to prioritize conservation efforts and monitor forest loss\n\n- Businesses: to meet “no deforestation” commitments, reduce supply-chain risk, and cut greenhouse gas emissions\n\nGlobal Recognition & Scalability\nThe approach aligns with the globally recognized HCS methodology and has been nominated for the UN SDG Game Changer award – Category 2: Planet – Innovation for our climate and environment “Artificial Intelligence Meets Jungle/ Forest Forward—a strong signal of technical credibility and market relevance. From Indonesia, the project adaptation scale on demand by Goa government, India and Senegal, West Africa.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Indonesia", "iso2": "ID" } ], "regions": [], "sdgs": [ "SDG 13", "SDG 15" ], "data_types": [ "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "CC-BY-SA 4.0", "spdx": "CC-BY-SA-4.0", "url": null }, "organizations": { "provided_by": { "text": "High Carbon Stock Approach (HCSA) Foundation, Jaringan Kerja Pemetaan Partisipatif (Seknas JKPP)", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All - GIZ, Bappenas - Ministry of State Development Planning (Indonesia)", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "JKPP (seknas@jkpp.org), HCSA (info@highcarbonstock.org), EcoVision Lab (jandirk.wegner@uzh.ch)", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: HCSA_Indonesia_Forest_Plot_Data_2…", "url": "https://huggingface.co/datasets/HCSA/HCSA_Indonesia_Forest_Plot_Data_2023" } ], "usecase": [ { "label": "highcarbonstock.org: mapping-portal", "url": "https://highcarbonstock.org/mapping-portal/" } ], "additional": [ { "label": "Ai Meets Jungle How Artificial Intelligence Is Saving Indonesias Forests (bmz-digital.global)", "url": "https://www.bmz-digital.global/en/news/ai-meets-jungle-how-artificial-intelligence-is-saving-indonesias-forests/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The finalized HCS dataset and maps are available on Hugging Face HCSA/Forest_Plot_Data_2023, and Global Forest Watch.\n \n • License: Open-source (CC BY 4.0) \n • Data Type: Geospatial and tabular \n • Key Variables: Vegetation classification, carbon stock levels, validation points, and FPIC (Free, Prior and Informed Consent) compliance\n • Appr. No of observations: Covers ~1.5 million hectares of tropical forests \n • Appr. No of dimensions: 60+ attributes, covering vegetation classifications, carbon stock levels, biomass measurements, land-use types, validation points, and FPIC (Free, Prior and Informed Consent) compliance data\n • Covering 150 field plot high-carbon stock areas across Sumatra, Kalimantan, and West Papua\n • Format: Geo TIFF, shapefile, CSV\n The data adheres to FAIR (Findable, Accessible, Interoperable, Reusable) principles and includes anonymization measures to protect sensitive information.\n\nAlso, for this dataset / use case, a responsible AI Assessment was undertaken to help AI developers and project managers to identify, assess and mitigate potential harms and biases in AI. For methodology, see https://www.bmz-digital.global/en/news/ethical-crash-test-for-ai-how-to-navigate-the-road-to-responsible-innovation/", "provenance": "curated" }, "model_characteristics": { "text": "Processed Data:\n Field measurements (e.g., biomass and tree species) were combined with Sentinel-2 imagery and NASA GEDI lidar to enhance map accuracy. The resulting geospatial data includes classifications of High Carbon Stock forests and other land-use categories.\n Model Development:\n AI models, trained using Convolutional Neural Networks (CNNs) and calibrated with aerial LiDAR and ground-truth data, are hosted on the HCSA GitHub repository: HCS Mapping Code.\n Application:\n Validated HCS maps are deployed on Google Earth Engine and Global Forest Watch for conservation planning and policy support, with an ODK-based mobile app under development for offline data collection. The project integrates FPIC (Free, Prior, and Informed Consent) principles throughout, ensuring community involvement and respect for land rights in data collection and mapping.", "provenance": "curated" }, "how_to_use": { "text": "Investment Opportunity\nInvestor support accelerates platform scaling, regional expansion, and integration with ESG reporting, carbon accounting, and supply-chain tools—unlocking long-term commercial and climate value.\n\nWhy Donor Support Matters\nFunding ensures continued data validation, open access, capacity building, and long-term sustainability—maximizing climate, biodiversity, and social returns. \n\nPath Forward\nGovernment collaboration enables institutionalization, national scaling, and integration into official planning and monitoring systems—supporting long-term, climate-aligned development.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_36/images/placeholder_image.jpg" }, { "id": "ui_37", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_37-building_inclusive_voice_technologies_in_the/", "aliases": [ "building_inclusive_voice_technologies_in_the", "text_data_collection_indonesia" ], "title": "Building inclusive voice technologies in the 3 Indonesian Languages Balinese, Bugis and Minangkabau through open language datasets", "description": { "text": "FAIR Forward and Prosa.ai collected AI training data and trained models for three digitally underrepresented languages of Indonesia: Balinese, Bugis and Minangkabau. Existing works on underrepresented languages mostly focus on machine translation and simple language understanding tasks, such as sentiment analysis and topic classification. More complex tasks such as open-domain dialogue and dialogue summarization, are still left behind for these underrepresented languages which leads to a poor evaluation suite for assessing the capability of large language models (LMs) in these underrepresented languages. Mitigating this limitation, we introduce NusaDialogue, the first underrepresented languages datasets consisting of manually annotated dialogue along with its summary.\n\nNusaDialogue covers 3 underrepresented languages under the Malayo-Polynesian languages group, i.e., Minangkabau (min), Balinese (ban), and Buginese (bug), each consisting of 10,000 dialogue and summary pairs. The dialogues in NusaDialogue encompass numerous topics including culture, occupation, politics, science, history, news, sports, religion, etc", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Indonesia", "iso2": "ID" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 10" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model", "tags": [ "dataset", "model" ] }, "license": { "name": "CC-BY-SA 4.0", "spdx": "CC-BY-SA-4.0", "url": null }, "organizations": { "provided_by": { "text": "Prosa.ai", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Prosa AI (business@prosa.ai)", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: nusa-dialogue", "url": "https://huggingface.co/datasets/prosa-text/nusa-dialogue" }, { "label": "Hugging Face: nusa-translation", "url": "https://huggingface.co/datasets/prosa-text/nusa-translation" } ], "usecase": [ { "label": "Hugging Face: indobert-nusa", "url": "https://huggingface.co/prosa-text/indobert-nusa" } ], "additional": [ { "label": "Entdeckung Der Sprachlichen Vielfalt Indonesiens Fair Forward Und Prosa Ai Auf Dem Weg Zu Inklusiver Ki Sprachtechnologie (bmz-digital.global)", "url": "https://www.bmz-digital.global/en/entdeckung-der-sprachlichen-vielfalt-indonesiens-fair-forward-und-prosa-ai-auf-dem-weg-zu-inklusiver-ki-sprachtechnologie/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "NusaDialogue covers 3 underrepresented languages under the Malayo-Polynesian languages group, i.e., Minangkabau (min), Balinese (ban), and Buginese (bug), each consisting of 10,000 dialogue and summary pairs. The dialogues in NusaDialogue encompass numerous topics including culture, occupation, politics, science, history, news, sports, religion, etc. \n\nThe non-translationese annotation nature makes NusaDialogue suitable for representing the actual day-to-day use of these underrepresented languages. In addition, we ensure that the dataset is annotated by a balanced number of male and female annotators to make the dataset represent a more balanced demography.\n\nAlso, a responsible AI Assessment was undertaken for this dataset / use case to help AI developers and project managers to identify, assess and mitigate potential harms and biases in AI. For methodology, see https://www.bmz-digital.global/en/news/ethical-crash-test-for-ai-how-to-navigate-the-road-to-responsible-innovation/", "provenance": "curated" }, "model_characteristics": { "text": "Pretrained Model: IndoBERT-Nusa, a BERT language model fine-tuned from indobert-large-p2 on Balinese, Buginese, and Minangkabau data. Framework: PyTorch / HuggingFace Transformers. License: CC-BY-SA 4.0. Tasks: fill-mask, topic classification, language identification. Topic classification F1 scores: Balinese 84.23, Buginese 82.03, Minangkabau 86.30. Training Data: NusaTranslation dataset (~140,000 sentence pairs per language, Indonesian to target language). Dialogue Data: NusaDialogue provides ~10,000 dialogue-summary pairs per language (3 million+ words total) across 16 topic categories, annotated by 100 qualified native speakers. Code for AI Models: load via HuggingFace (`prosa-text/indobert-nusa`).\n\nSource: https://huggingface.co/datasets/prosa-text/nusa-dialogue, https://huggingface.co/datasets/prosa-text/nusa-translation, https://huggingface.co/prosa-text/indobert-nusa", "provenance": "auto-enriched" }, "how_to_use": { "text": "Supporting Minority Languages in the Digital Space:\n\nThe digital divide is often more pronounced for speakers of minority and underrepresented languages. Including these languages in summarization datasets helps bridge this gap, ensuring that digital platforms and technologies are accessible and beneficial to a broader range of linguistic communities.\n\nEncouraging Linguistic Research\n\nThe creation of NusaDialogue can stimulate interest and research in linguistics, language preservation, and computational linguistics. This can lead to a better understanding of the unique linguistic features of these languages and the development of more tailored language technologies.\n\nAddressing Bias and Fairness in NLP\n\nThrough NusaDialogue , there is an opportunity to address biases present in NLP systems that are often trained on data from dominant languages. This can contribute to the development of more fair and unbiased language technologies.\n\nIn conclusion, the social impact of NusaDialogue extends beyond technology to cultural preservation, education, inclusivity, and cross-cultural understanding. It has the potential to empower communities, promote linguistic diversity, and contribute to a more equitable and interconnected global society.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_37/images/placeholder_image.jpeg" }, { "id": "ui_38", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_38-empowering_coastal_inhabitants_in_indonesia_levera/", "aliases": [ "blue_economy_for_climate_mitigation_and", "empowering_coastal_inhabitants_in_indonesia_levera" ], "title": "Empowering coastal inhabitants in Indonesia: leveraging AI, community-based approaches and local Weather measurements for enhanced climate adaptation and a thriving “blue economy”", "description": { "text": "This dataset helps researcher observe daily weather changes, analyze local climate patterns, and support research, planning, or environmental modeling through machine learning and data science in Balai Silvofishery Maros Regency, South Sulawes. Weather Data is a collection of environmental condition records gathered from a monitoring point located in Balai Silvofishery Maros Regency. The dataset includes key weather parameters such as temperature, humidity, rainfall, light intensity, wind speed, and other atmospheric indicators. This data can enable accurate weather predictions which are critical for small-scale fishing, ecotourism, and coastal protection. By empowering fishermen and women with real-time insights, the dataset can serve to build systems that enhance safety, optimise livelihoods, and support sustainable local, blue economies.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Indonesia", "iso2": "ID" } ], "regions": [], "sdgs": [ "SDG 13", "SDG 14" ], "data_types": [ "Meterological" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": null, "organizations": { "provided_by": { "text": "Common Room Networks Foundation", "links": [ { "name": "Common Room Networks Foundation", "url": "https://commonroom.info/" } ] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Common Room Networks Foundation (email.commonroom@gmail.com)", "access_note": null, "links": { "dataset": [ { "label": "GitHub: WeatherData", "url": "https://github.com/insaninfonesia/WeatherData" } ], "usecase": [ { "label": "colabs.commonroom.info: login", "url": "https://colabs.commonroom.info/login/?next=%2F" } ], "additional": [ { "label": "Ai Climate Resilience Meets Local Know How (giz.de)", "url": "https://www.giz.de/en/newsroom/stories/ai-climate-resilience-meets-local-know-how" } ], "documents": [] }, "content": { "data_characteristics": { "text": "License: CC-BY-SA 4.0. Data Type: Environmental monitoring records. Locations: Balai Silvofishery, Maros Regency, South Sulawesi Province and Pulo Aceh, Aceh Besar Regency, Aceh Province, Indonesia. Key Variables: temperature, humidity, rainfall, light intensity, wind speed. Collection period: since August 2025 (~7 months of data). Format: compressed archives (WS_MAROS_April-November2025.zip, WS_PULO_ACEHwoNew.txz). Known data quality issues: pressure and wind speed values entirely absent; temperature constant at 27.7C; humidity around 80%; irregular data intervals; Pulo Aceh data incomplete due to power outages and limited internet. Contact: infonesiainsan@gmail.com.\n\nSource: https://github.com/insaninfonesia/WeatherData", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Model: https://colabs.commonroom.info/login/?next=%2F with username: guest_viewer and the password: guest_viewer", "provenance": "curated" }, "how_to_use": { "text": "This data and AI models built on the data can enable systems that provide accurate weather predictions which are critical for small-scale fishing, ecotourism, and coastal protection. By empowering fishermen and women with real-time insights, such systems can enhance safety, optimise livelihoods, and support sustainable local, blue economies.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_38/images/placeholder_image.jpeg" }, { "id": "ui_40", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_40-early_warning_system_advisory_services_on/", "aliases": [ "early_warning_system_advisory_services_on", "early_warning_system_ldri" ], "title": "Early Warning System: Advisory services on climate-smart farming for small-holder farmers", "description": { "text": "The Early Warning System (EWS) is an AI-powered platform that monitors farming activities and supports climate-smart precision agriculture for smallholder farms. It uses geo-referenced data collected from farms through mobile phones and high-resolution satellite imagery. The system provides crop monitoring, yield prediction, pest/disease detection, and localized agricultural advisory services in multiple languages. The platform is currently accessed through both WhatsApp by farmers and a self-serve platform/API for institutional users.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Kenya", "iso2": "KE" } ], "regions": [], "sdgs": [ "SDG 10", "SDG 2" ], "data_types": [ "Geospatial/Remote Sensing", "Images", "Text" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": null, "organizations": { "provided_by": { "text": "Local Development Research Institute (LDRI)", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "LDRI (thinking@developlocal.org)", "access_note": null, "links": { "dataset": [ { "label": "Google Drive: 19ZMqreP_lWXE13gbTGcdgYFlNSXXB9AN", "url": "https://drive.google.com/drive/u/0/folders/19ZMqreP_lWXE13gbTGcdgYFlNSXXB9AN" } ], "usecase": [ { "label": "vuli.ai", "url": "https://www.vuli.ai/" } ], "additional": [ { "label": "Smart Sowing Ai For More Harvest In Kenya And Uganda (bmz-digital.global)", "url": "https://www.bmz-digital.global/en/smart-sowing-ai-for-more-harvest-in-kenya-and-uganda/" }, { "label": "Food Nutrition Early Warning Mechanism (developlocal.org)", "url": "https://www.developlocal.org/food-nutrition-early-warning-mechanism/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Ground truth geo-tagged data from 1,521 farms across 5 regions in Kenya\n Types of data: Crop images, farm boundaries, crop conditions, weekly activity logs\n Coverage: Multiple crop types including maize, beans, sweet potatoes, cassava, kales, Irish potatoes, cowpeas, soya beans, and indigenous vegetables\n Language data: ~150 hours of voice data and transcriptions in KiEmbu, Kikuyu, Kiswahili and Luhya\n Timeline: Data collected over multiple growing seasons from 2022-2024\n Collection conducted by LDRI in partnership with champion farmers and agronomists.", "provenance": "curated" }, "model_characteristics": { "text": "Processed Data and Models:\n 1. Uses geo-tagged images collected from farms and annotated for crop types, health status, and specific afflictions (pests/diseases/nutrition deficiency) \n 2. Derived vegetation indices, moisture indices, and time-series data from satellite imagery.\n 3. A corpus of voice and text data is being developed for local language processing.\n 4. Crop classification models using CNN architectures : Identifies \n 5. Crop Yield prediction models \n 6. Language models for agricultural advisory in multiple local languages (in development)\n Application:\n The system can be used for:\n • Early warning of potential crop failure\n • Yield prediction and crop health monitoring\n • Localized agricultural advisory in multiple languages\n • Farm productivity tracking and management\n • Regional agricultural monitoring and planning", "provenance": "curated" }, "how_to_use": { "text": "The Eearly Warning System provides a comprehensive platform that combines AI-powered crop monitoring with localized advisory services. Organizations can access the data and models through :A self-serve platform and API for accessing predictions and analytics; Mobile applications for farmers to receive personalized crop management recommendations in local language; Integration possibilities with existing agricultural extension services and Analytics dashboards for value chain stakeholders and policy institutions. The system can be scaled to new regions and crops through transfer learning and the modular architecture allowing for continuous improvement of individual components. The platform supports both individual farmer-level insights and aggregated regional analytics to optimize crop yields and manage stressors effectively, making it valuable for various stakeholders across the agricultural sector.\nThis use case is also linked to the development of a business model / funding model for open source AI as a stepping stone towards financially viable operations. It was part of Villgro Africa's and \"FAIR Forward — AI for All\"'s six-month mentorship programme on \"Creating sustainable business and funding models with open source AI\". See: https://www.bmz-digital.global/en/news/open-source-ai-business-impact/", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_40/images/placeholder_image.jpeg" }, { "id": "ui_42", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_42-paza_sauti__chatbot_and_ivr/", "aliases": [ "paza_sauti__chatbot_and_ivr", "support_mozilla_swahili_grantee__part" ], "title": "Paza Sauti - chatbot and IVR service in Swahili to raise awareness about the use of collateral (security) to access credit for women in Kenya", "description": { "text": "The project is developing a chatbot and an interactive voice response service that will provide voice-enabled services in the domain of business registration and raise awareness about the use of collateral (security) to access credit in Kenya. The main objective is to increase financial literacy around moveable properties as collateral, particularly for women in business, and in particular agriculture, for purposes of accessing credit. Although there has been an increase in the ease of getting credit, most members of the population are still unaware of their capability to access further credit as a result of using moveable properties as collateral. This was part of a Mozilla innovation challenge supporting people and projects across East Africa who leverage Common Voice’s open-source voice data set to unlock social and economic opportunities. These grants help to advance the use of open-source voice data for products that support community participation and engagement.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Kenya", "iso2": "KE" } ], "regions": [], "sdgs": [ "SDG 10", "SDG 5", "SDG 2" ], "data_types": [ "Text", "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": null, "organizations": { "provided_by": { "text": "Tech Innovators Network Kenya (THiNK)", "links": [] }, "catalyzed_by": { "text": "Mozilla Foundation & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "Gates Foundation & BMZ", "links": [] } }, "contact": "THiNK (support@think.ke)", "access_note": null, "links": { "dataset": [ { "label": "GitHub: sheng-dataset", "url": "https://github.com/think-ke/sheng-dataset/" } ], "usecase": [ { "label": "GitHub: mpsr", "url": "https://github.com/think-ke/mpsr" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": null, "model_characteristics": null, "how_to_use": null }, "image": "https://fair-forward.github.io/datasets/projects/ui_42/images/placeholder_image.jpeg" }, { "id": "ui_47", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_47-aipowered_livestock_health_system_enabling_local/", "aliases": [ "aipowered_livestock_health_system_enabling_local" ], "title": "AI-powered livestock health system enabling local communities easy access to disease information on demand and in Kiswahili.", "description": { "text": "LivHealth Kiswahili Corpus aims to empower local communities to correctly identify livestock syndromes and get timely interventions from qualified livestock practitioners. Using Natural Language Processing (NLP), Machine Learning (ML), and Artificial Intelligence (AI), the project has built Kiswahili text-to-speech models for disseminating disease information to marginalized communities. Working closely with their partner, One Health Center in Africa (OHRECA) based at ILRI, they have enhanced the functionality of the LivHealth system to enable local communities easy access to disease information on demand and in Kiswahili. This was part of a Mozilla innovation challenge supporting people and projects across East Africa who leverage Common Voice’s open-source voice data set to unlock social and economic opportunities. These grants help to advance the use of open-source voice data for products that support community participation and engagement.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Kenya", "iso2": "KE" } ], "regions": [ "East Africa" ], "sdgs": [ "SDG 10", "SDG 2" ], "data_types": [ "Text", "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": null, "organizations": { "provided_by": { "text": "Badili Innovations", "links": [] }, "catalyzed_by": { "text": "Mozilla Foundation & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "Gates Foundation & BMZ", "links": [] } }, "contact": "Badili Innovations (info@badili.co.ke)", "access_note": null, "links": { "dataset": [ { "label": "livhealth-kc.badili.co.ke", "url": "https://livhealth-kc.badili.co.ke/" } ], "usecase": [ { "label": "livhealth-kc.badili.co.ke", "url": "https://livhealth-kc.badili.co.ke/" } ], "additional": [ { "label": "New App And Local Language Translations Improve Livestock Health In Kenya And Tanzania (foundation.mozilla.org)", "url": "https://foundation.mozilla.org/blog/new-app-and-local-language-translations-improve-livestock-health-in-kenya-and-tanzania/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Data Type: Text, Voice. Language: Kiswahili. Country: Kenya. The corpus contains livestock disease information translated from English to Kiswahili, including disease descriptions, symptoms, and treatment protocols. Data is collected via a smartphone app used by Community Disease Reporters (CDRs) and pastoralists. Maintained by Badili Innovations in partnership with One Health Center in Africa (OHRECA) at ILRI.\n\nSource: https://livhealth-kc.badili.co.ke/", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Application: LivHealth uses NLP and ML to provide Kiswahili text-to-speech output of livestock disease information, enabling communities with literacy challenges to access disease identification and treatment guidance. Users input livestock symptoms and receive spoken Kiswahili responses. Requires an internet-connected device with audio playback. Built by Badili Innovations in partnership with OHRECA at ILRI. Part of a Mozilla Common Voice innovation challenge supporting open-source voice technology in East Africa.\n\nSource: https://livhealth-kc.badili.co.ke/", "provenance": "auto-enriched" }, "how_to_use": { "text": "LivHealth is a web-based knowledge platform that makes livestock health information accessible in Kiswahili. Developed by Badili Innovations in partnership with the International Livestock Research Institute (ILRI), it is designed for livestock practitioners, extension workers, and development organisations operating in East African Swahili-speaking communities.\n\nYou can access the platform directly at livhealth-kc.badili.co.ke to browse structured livestock health knowledge in Kiswahili. This makes it a practical resource for field programmes that need locally relevant, language-appropriate veterinary and animal health information -- whether for training community animal health workers, supporting extension services, or informing livestock health interventions.\n\nThe platform spans three functional areas: information systems for structured livestock health knowledge, data management covering collection, synthesis, analytics and visualisation, and monitoring, evaluation and learning (MEL) to support tracking of livestock health interventions.\n\nFor questions about integrating LivHealth into livestock health programmes or extending the platform, contact the development team through Badili Innovations at badili.co.ke. ILRI provides the domain expertise on livestock health content.\n\nSource: https://livhealth-kc.badili.co.ke/, https://badili.co.ke", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_47/images/placeholder_image.jpeg" }, { "id": "ui_53", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_53-detecting_sentiments_and_combatting_hate_speech/", "aliases": [ "a_nigerian_twitter_sentiment_corpus_for", "detecting_sentiments_and_combatting_hate_speech" ], "title": "Detecting sentiments and combatting hate speech in Hausa, Igbo, Nigerian-Pidgin and Yorùbá - NaijaSenti: a Nigerian Corpus for Multilingual Sentiment Analysis", "description": { "text": "The NaijaSenti dataset is the first large-scale human-annotated Twitter sentiment dataset for Hausa, Igbo, Nigerian-Pidgin, and Yorùbá, the four most widely spoken languages in Nigeria. It consists of around 30,000 annotated tweets per language (except for Nigerian-Pidgin), including a significant fraction of code-mixed tweets. These datasets are useful not only for sentiment analysis but also for hate speech detection. The open-source package consists of datasets, trained models, sentiment lexicons, and code. With more than 200 million people and 522 native languages, Nigeria is the most populous and linguistically diverse country in Africa, as well as the third most multilingual country in the world. The majority of the population speaks either Hausa, Igbo, Yorùbá, or Nigerian-Pidgin.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Nigeria", "iso2": "NG" } ], "regions": [], "sdgs": [ "SDG 10" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Bayero University (Kano, Nigeria), Ahmadu Bello University (Zaria, Nigeria), Masakhane NLP, HausaNLP, Kaduna State University", "links": [] }, "catalyzed_by": { "text": "Lacuna-Fund / Meridian (Climate-call) & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "NaijaSenti / HausaNLP (hausanlp@gmail.com)", "access_note": null, "links": { "dataset": [ { "label": "GitHub: NaijaSenti", "url": "https://github.com/hausanlp/NaijaSenti" } ], "usecase": [ { "label": "Hugging Face: naija-twitter-sentiment-afriberta…", "url": "https://huggingface.co/Davlan/naija-twitter-sentiment-afriberta-large" } ], "additional": [ { "label": "2201.08277 (arxiv.org)", "url": "https://arxiv.org/pdf/2201.08277" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The project developed and made the following resources available: 1. Manually Annotated Twitter Sentiment Dataset 2. Manually Annotated Sentiment Lexicon 3. Semi-automatically Translated emotion lexicon 4. Semi-automatically Translated sentiment lexicon 5. Large Scale Unlabeled Twitter Sentiment Corpus 6. Stop-words for Hausa, Igbo, Pidgin and Yoruba 7. text collection, filtering, processing and labeling methods to create datasets for these low-resource languages. 8. evaluation og a range of pre-trained models and transfer strategies on the dataset. 9. release of datasets, trained models, sentiment lexicons, and code to incentivize research on sentiment analysis in under-represented languages. Authors: Shamsuddeen Hassan Muhammad, David Ifeoluwa Adelani, Sebastian Ruder, Ibrahim Said Ahmad, Idris Abdulmumin, Bello Shehu Bello, Monojit Choudhury, Chris Chinenye Emezue, Saheed Salahudeen Abdullahi, Anuoluwapo Aremu, Alipio Jeorge, and Pavel Brazdil", "provenance": "curated" }, "model_characteristics": null, "how_to_use": { "text": "NaijaSenti can be used to advance sentiment analysis and other downstream NLP tasks in the languages involved and to work on hate speech detection. For such follow-up work on hate speech, see also: https://arxiv.org/abs/2501.08284", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_53/images/sentiment.pn.jpg" }, { "id": "ui_55", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_55-landslide_and_flood_disaster_hotspot_monitoring/", "aliases": [ "landslide_and_flood_disaster_hotspot_monitoring", "landslide_monitoring_use_case", "onsite_image_and_weather_data_for" ], "title": "Landslide and flood disaster hotspot monitoring using computer vision in Rwanda", "description": { "text": "The iMaster-DocuCam Landslide Monitoring System by Hesotech GmbH provides long-term, continuous visual documentation of landslide-prone areas in Rwanda. A pan-tilt-zoom (PTZ) camera captures high-resolution images (up to 1 cm per cell at 500 m distance) at user-defined intervals, while supplementary sensors record temperature, rainfall, humidity, and GPS data. All data is stored in a single database and accessible via web browser from any location. The system is designed for low energy consumption and can be powered by solar panels or generators, making it suitable for remote areas. A demo version and online user manual are available at https://manual.docucam.hesotech.eu/en.\n\nSource: https://www.hesotech.de/en/products-services/imaster-docucam-landslide-monitoring-system/", "provenance": "auto-enriched" }, "kind": [ "usecase" ], "countries": [ { "name": "Rwanda", "iso2": "RW" } ], "regions": [], "sdgs": [ "SDG 13", "SDG 11" ], "data_types": [ "Images", "Meterological" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Hesotech GmbH", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Hesotech (info@hesotech.de)", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "hesotech.de: imaster-docucam-landslide-monitor…", "url": "https://www.hesotech.de/en/products-services/imaster-docucam-landslide-monitoring-system/" } ], "additional": [ { "label": "Papers.Cfm (papers.ssrn.com)", "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5087469" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The system captures high-resolution images (up to 1 cm per cell at 500 m distance) at predefined intervals alongside environmental data (temperature, rainfall, humidity, GPS). Data is stored in a single database and accessible via web browser. The hardware consists of a PTZ camera and electronics cabinet with mini-PC; optional expansions include a local weather station, LTE/3G/4G connection, antenna, generator, solar panel, and custom sensor interfaces. Connectivity via local WiFi or LTE/3G/4G. No physical markers required in the monitored area; virtual markers can be created on-screen. Supports time-lapse generation, built-in ruler measurements, and image annotation (text, shapes, arrows, pixelation of sensitive areas).\n\nSource: https://www.hesotech.de/en/products-services/imaster-docucam-landslide-monitoring-system/", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Performs long-term, continuous visual monitoring to detect small and large, slow and fast surface changes for landslide and earth movement analysis. Automated image capture at user-defined intervals with simultaneous collection of weather and GPS data enables cause-effect correlation. Does not require physical markers; users create virtual markers on-screen. Generates time-lapse recordings to visualize changes over time. Web-browser-based access from any location. Low energy consumption allows operation via solar panel or generator in off-grid areas. Developed by Hesotech GmbH.\n\nSource: https://www.hesotech.de/en/products-services/imaster-docucam-landslide-monitoring-system/", "provenance": "auto-enriched" }, "how_to_use": { "text": "## How to Use This Resource\n\nThe iMaster-DocuCam system by Hesotech enables long-term, high-resolution visual monitoring of landslides, earthfalls, rockfalls, and earth movements across large areas. It is designed for deployment in environments like Rwanda where continuous observation of hazard-prone slopes is critical for disaster risk reduction.\n\nPractitioners working in disaster risk management, climate adaptation, or infrastructure protection can use this system to establish automated visual monitoring stations that capture images at predefined intervals (for example, every two hours), achieving a resolution of up to 1 cm per cell at 500 m distance. All captured data can be viewed remotely through a standard web browser (Chrome or Firefox) from any location -- no proprietary software is required. This makes the system practical for distributed teams and remote field sites.\n\nThe built-in analysis tools allow you to place virtual measurement points on images without physical markers in the field, measure surface changes over time, generate time-lapse videos to visualise movement patterns, and correlate camera data with complementary sensor readings for cause-effect analysis. The system can be extended with a local weather station, GPS data collection, custom sensor interfaces, and solar panel power supply for off-grid deployment, making it suitable for rural and infrastructure-limited contexts.\n\nA comprehensive online manual with animated tutorials is available at [manual.docucam.hesotech.eu/en](https://manual.docucam.hesotech.eu/en), along with a demo version that allows you to explore the interface before committing to a deployment. Full product specifications and contact details can be found on the [Hesotech product page](https://www.hesotech.de/en/products-services/imaster-docucam-landslide-monitoring-system/).\n\nSource: https://www.hesotech.de/en/products-services/imaster-docucam-landslide-monitoring-system/", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_55/images/placeholder_image.jpeg" }, { "id": "ui_56", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_56-mbaza_chatbot_rwanda_for_health_related/", "aliases": [ "develop_mbaza_ai_chatbot_for_covid", "mbaza_chatbot_rwanda_for_health_related" ], "title": "Mbaza Chatbot Rwanda for health related inquiries by citizens", "description": { "text": "The Mbaza AI Chatbot was awarded as one of the winning projects of the #SmartDevelopmentHack, an international hackathon organized by the German Federal Ministry for Economic Cooperation and Development (BMZ). The hackathon called for innovative digital solutions to tackle the challenges caused by the coronavirus outbreak in low- and middle-income countries. The chatbot was developed iteratively, the USSD channel was the first to go production, while the web and IVR versions reached only staging phase. Important reported metrics for the Mbaza covid-19 ussd chatbot: 4,165,037 users in total\n37,812,089 sessions in total\n7,254,441 sessions in January 2022\n1,337,614 users in January 2022\n4,347,713 session in December 2021\n914,530 users in December 2021 \n10,074 users in September 2024\n69,992 sessions in September 2024\n170,857 users in 2024\n1,073,451 sessions in 2024", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Rwanda", "iso2": "RW" } ], "regions": [], "sdgs": [ "SDG 10" ], "data_types": [ "Text", "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": { "name": "CC0 1.0", "spdx": "CC0-1.0", "url": null }, "organizations": { "provided_by": { "text": "Digital Umuganda", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ, Digital Transformation Centre Rwanda", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Digital Umuganda, Samuel Rutunda (samuel@digitalumuganda.com), Audace Niyonkuru", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: common-voice-kinyarwanda-text-dat…", "url": "https://huggingface.co/datasets/DigitalUmuganda/common-voice-kinyarwanda-text-dataset" }, { "label": "Hugging Face: common-voice-kinyarwanda-english-…", "url": "https://huggingface.co/datasets/mbazaNLP/common-voice-kinyarwanda-english-dataset" }, { "label": "datacollective.mozillafoundation.org", "url": "https://datacollective.mozillafoundation.org/" } ], "usecase": [ { "label": "GitHub: Mbaza-chatbot", "url": "https://github.com/Digital-Umuganda/Mbaza-chatbot" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "The tabular dataset for the RASA model can be requested from the startup Digital Umuganda or the implementing partner RBC (as this was specific to daily statistics, guidelines, and announcements for pandemic prevention published by the authority). The Common Voice datasets can be found under the dataset links:", "provenance": "curated" }, "model_characteristics": { "text": "Conversational chatbot built on the Rasa open-source framework, providing COVID-19 information to Rwandan citizens in English and Kinyarwanda. Covers outbreak statistics, symptoms, prevention tips, fines/penalties, testing center locations, testing procedures, costs, lockdown areas, and vaccination information. Developed by Digital Umuganda. The USSD channel reached production; web and IVR versions reached staging phase. Reported usage: 4,165,037 total users and 37,812,089 total sessions. Source code available on GitHub.\n\nSource: https://github.com/Digital-Umuganda/Mbaza-chatbot", "provenance": "auto-enriched" }, "how_to_use": { "text": "This work can be used by anyone interested to build a conversational chatbot in English and Kinyarwanda that needs to provide key information during disease outbreaks such as checking outbreak statistics, showing some prevention tips, checking testing centers, vaccination information etc.\n\nThis use case also included the development of a business model and funding model for open source AI as a stepping stone towards financially viable operations. It was facilitated by Villgro Africa's and FAIR Forward's six-month mentorship programme on \"Creating sustainable business and funding models with open source AI\". See: https://www.bmz-digital.global/en/news/open-source-ai-business-impact/", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_56/images/placeholder_image.jpeg" }, { "id": "ui_60", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_60-solutions_from_space_rwandas_smart_harvest/", "aliases": [ "ml4eo_crop_mapping_solution_rwanda", "solutions_from_space_rwandas_smart_harvest" ], "title": "Solutions from space: Rwanda’s smart harvest planning - mapping crop type with AI for rice, maize, Irish potatoes and beans", "description": { "text": "When weather and crop yields become unpredictable, reliable information is crucial. In Rwanda, digital maps are showing for the first time exactly where which crops grow, helping to better adapt agriculture to climate change. Around one third of all food worldwide is lost between the field and the plate, for example due to pest infestation, incorrect storage or long transport routes. Climate change is exacerbating these losses: irregular rainfall, dry seasons and extreme weather events are jeopardising harvests and making planning increasingly difficult for farmers. Rwanda’s agricultural sector also faces these challenges. \n\nThis automated crop type mapping system for Rwanda can be used in various applications e.g. improving agricultural policy and business decisions, integrating with yield estimation frameworks for spatially detailed predictions of seasonal production, crop health monitoring, crop insurance design, value chain assessments, supply and demand forecasting, and more.\n\nIt leverages multi-sensor remote sensing data (Sentinel 1 & 2, dynamic world, SRTM). The crop mapping pipeline can be used to map the four most important staple foods in Rwanda (rice, maize, beans, Irish potatoes) in four Rwandan districts, and produce timely, 10-meter resolution crop type maps. \nReference data from different sources including project field work, global research projects, and Rwandan authorities were used to fine-tune the geospatial foundational model Galileo for the downstream task of mapping what type of crop grows where. The reference data collected in the field was supported through the creation of polygon labels with PlanetScope imagery in collaboration with Rwandan academia. The crop mapping pipeline, the fine-tuned model, and the project's reference data are openly available for further usage.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Rwanda", "iso2": "RW" } ], "regions": [], "sdgs": [ "SDG 2" ], "data_types": [ "Geospatial/Remote Sensing" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": { "name": "MIT", "spdx": "MIT", "url": null }, "organizations": { "provided_by": { "text": "Rwanda Space Agency (RSA), the Rwandan Ministry of Agriculture, Alliance of Bioversity International and CIAT", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All (GIZ), Digital Transformation Centre Rwanda", "links": [] }, "financed_by": { "text": "BMZ", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Benson Kenduiyvo (b.kenduiywo@cgiar.org)", "access_note": null, "links": { "dataset": [ { "label": "dataverse.harvard.edu: dataset.xhtml", "url": "https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/O7IDTD" } ], "usecase": [ { "label": "GitHub: crop-type-mapping", "url": "https://github.com/crop-type-mapping" } ], "additional": [ { "label": "Digital Maps Agriculture Rwanda (bmz-digital.global)", "url": "https://www.bmz-digital.global/en/digital-maps-agriculture-rwanda/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The overall good accuracy yielded by the approach under conditions with small field sizes, varying elevation and high cloud cover are promising for a successful application in all kinds of regions. Nevertheless, local reference data remains crucial for crop type mapping and an ideal approach should include all types of crops. You can follow the approach to create your own crop type mapping system, utilize the pipeline to fetch satellite imagery for various purposes, get inspired by the use and fine-tuning of the geospatial foundational model and apply it to a different task. You can also use the collected reference data for your own model.", "provenance": "curated" }, "model_characteristics": { "text": "Machine learning pipeline for crop type classification in Rwanda using Sentinel-1 and Sentinel-2 satellite imagery. Training scripts include Random Forest and climate-corrected models. Ground reference data covers four districts (Nyagatare, Musanze, Nyabihu, Ruhango) with field observations of maize, beans, rice, and Irish potato collected May-June 2025. The ground truth dataset contains GPS-located farm plots with crop attributes, management practices, and directional photographs. Code: Python scripts for satellite data preparation, image download, and model training at github.com/crop-type-mapping/ML. License: MIT (code), CC-BY 4.0 (ground reference data).\n\nSource: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/O7IDTD, https://github.com/crop-type-mapping", "provenance": "auto-enriched" }, "how_to_use": { "text": "The mapping method developed is freely accessible and fully documented. Training courses with the Rwandan Space Agency and the Office of Statistics ensure that local experts can continue to develop the system independently.\r\nThe digital maps do not replace observations and measurements in the field or local knowledge. They complement these to create an improved basis for data-driven decisions, more targeted use of resources and forward-looking risk management using digital tools – for agriculture that offers prospects for the future even under climate pressure.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_60/images/placeholder_image.jpeg" }, { "id": "ui_61", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_61-better_language_translation_for_more_training/", "aliases": [ "better_language_translation_for_more_training" ], "title": "Better language translation for more training content in Kinyarwanda - education-specific machine translation for the Moodle Learning Management System", "description": { "text": "Enabling language translation capabilities on the Moodle LMS platform, through a collaboration with Atingi; the use case explores 3 modes of translation. The first mode is navigation based, simply translating static messages such as buttons and notifications, secondly it explores full content translation, lastly, in-line translation. The project uses the NLLB-200 model for translation, other models could be experimented to explore feasibility of additional features such as QA or summarization, leveraging the plugin built by the team.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Rwanda", "iso2": "RW" } ], "regions": [], "sdgs": [ "SDG 4" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "Permissive", "spdx": null, "url": null }, "organizations": { "provided_by": { "text": "Digital Umuganda", "links": [] }, "catalyzed_by": { "text": "Atingi, \"FAIR Forward - AI for All\" (GIZ)", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Digital Umuganda (samuel@digitalumuganda.com), FAIR Forward Rwanda", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: kinyarwanda-english-machine-trans…", "url": "https://huggingface.co/datasets/DigitalUmuganda/kinyarwanda-english-machine-translation-dataset" }, { "label": "Hugging Face: Kinyarwanda_English_parallel_data…", "url": "https://huggingface.co/datasets/mbazaNLP/Kinyarwanda_English_parallel_dataset" }, { "label": "Hugging Face: NMT_Education_parallel_data_en_kin", "url": "https://huggingface.co/datasets/mbazaNLP/NMT_Education_parallel_data_en_kin" } ], "usecase": [ { "label": "Hugging Face: Quantized_Nllb_Finetuned_Edu_En_K…", "url": "https://huggingface.co/mbazaNLP/Quantized_Nllb_Finetuned_Edu_En_Kin_8bit" }, { "label": "Hugging Face: Nllb_finetuned_general_en_kin", "url": "https://huggingface.co/mbazaNLP/Nllb_finetuned_general_en_kin" }, { "label": "Hugging Face: Finetuned-NLLB", "url": "https://huggingface.co/DigitalUmuganda/Finetuned-NLLB" } ], "additional": [ { "label": "Moodlemoot Global Day 3 (moodle.com)", "url": "https://moodle.com/news/moodlemoot-global-day-3/" }, { "label": "aiopeneducation.pubpub.org", "url": "https://aiopeneducation.pubpub.org/pub/7npwm2su/release/5?readingCollection=06969c6d" } ], "documents": [] }, "content": { "data_characteristics": { "text": "Parallel datasets can be used to finetune LLMs to specific languages or domains; the example is for English-Kinyarwanda, to improve NLLB translation performance", "provenance": "curated" }, "model_characteristics": { "text": "Open source model NLLB-200, some variations quantized to minimize compute resource usage", "provenance": "curated" }, "how_to_use": { "text": "Integrate preferred machine translation model with Moodle plugin: https://github.com/Digital-Umuganda/moodle-translate_courses . Need to update for compatibility with newer Moodle updates, this was implemented during Moodle version 3.0", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_61/images/placeholder_image.jpeg" }, { "id": "ui_75", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_75-crop_type_identification_from_satellite_imagery/", "aliases": [ "crop_type_identification_from_satellite_imagery", "south_africa_crop_type_competition", "spot_the_crop" ], "title": "Crop Type Identification from Satellite Imagery in Western Cape, South Africa", "description": { "text": "This dataset and AI model was produced as part of the Radiant Earth Spot the Crop Challenge (https://zindi.africa/hackathons/radiant-earth-spot-the-crop-hackathon). The objective of the competition was to create a machine learning model to classify fields by crop type from images collected during the growing season by the Sentinel-2 and Sentinel-1 satellites.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "South Africa", "iso2": "ZA" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 13" ], "data_types": [ "Tabular" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Radiant Earth, Western Cape Department of Agriculture", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Radiant Earth (ml@radiant.earth)", "access_note": null, "links": { "dataset": [ { "label": "beta.source.coop: description", "url": "https://beta.source.coop/repositories/radiantearth/south-africa-crops-competition/description/" } ], "usecase": [ { "label": "GitHub: spot-the-crop-challenge", "url": "https://github.com/radiantearth/spot-the-crop-challenge" } ], "additional": [ { "label": "Radiant Earth Spot The Crop Challenge (zindi.africa)", "url": "https://zindi.africa/competitions/radiant-earth-spot-the-crop-challenge" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The dataset contains a time-series of satellite imagery and labels for crop type that have been collected through aerial and ground survey. Labels are derived from the survey conducted by the Western Cape Department of Agriculture. Satellite data including multispectral Sentinel-2 are then matched with corresponding labels.", "provenance": "curated" }, "model_characteristics": { "text": "This repository contains the winning models from the Zindi data competition \"Spot the Crop\" in which participants used time series of Sentinel-2 multispectral imagery as input for crop type classification.", "provenance": "curated" }, "how_to_use": { "text": "The agricultural sector makes a substantial contribution to GDP and livelihoods across the developing world. However, regular and reliable agricultural data remains difficult and expensive to collect on the ground. As a result, policy-makers usually don’t have access to updated data for implementing policies or supporting farmers. Earth observation satellites provide a wealth of multispectral image data that can be used for developing agricultural monitoring tools. These tools support farmers and policy-makers across Africa and the developing world. With this dataset and AI model you can use time-series of Sentinel-2 multi-spectral data to classify crops in the Western Cape of South Africa.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_75/images/crop.png" }, { "id": "ui_78", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_78-croppie_helping_smallholder_coffee_producers_to/", "aliases": [ "croppie_coffee_yield_prediction", "croppie_helping_smallholder_coffee_producers_to" ], "title": "Croppie- helping smallholder coffee producers to plan sales, estimate yields, get loans, and trace coffee - AI powered coffee yield prediction", "description": { "text": "Contributing almost a third of foreign export earnings, coffee is one of the main cash crops in Uganda. Nowadays, many small-holder farmers, who rely on coffee farming for their livelihood, are facing challenges such as unpredictable weather and volatile market prices. Traditionally, they have relied on manual yield estimates, but these are time-intensive to perform and not always reliable, leaving many farmers unable to safely plan their harvests or manage their resources effectively. On top, unpredictable weather conditions and price fluctuations pose major challenges for the coffee farmers. \r\n\r\nThe platform and app 'Croppie' supports them with targeted cultivation recommendations and precise crop forecasts—using AI-supported image recognition of coffee plants. As a digital public good based on open-source technology, it can be adapted and further developed locally. The basic user interface is provided through a mobile application. Croppie’s core functionality is to allow for AI-supported coffee yield estimation. This is done by guiding a selective sampling for a statistical yield estimation service based on branch count and coffee branch picture taking, alongside the integration of additional data provided by the user.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Uganda", "iso2": "UG" } ], "regions": [], "sdgs": [ "SDG 2", "SDG 13" ], "data_types": [ "Tabular" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": { "name": "CC-BY-SA 4.0", "spdx": "CC-BY-SA-4.0", "url": null }, "organizations": { "provided_by": { "text": "Producers Direct, M-Omulimisa, Alliance of Bioversity International and CIAT, Tecnicafé", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Croppie - Romain Gautron (r.gautron@cgiar.org), Producers Direct (info@producersdirect.org), M-Omulimisa, CIAT", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: croppie_coffee_ug", "url": "https://huggingface.co/datasets/rgautroncgiar/croppie_coffee_ug" } ], "usecase": [ { "label": "Hugging Face: croppie_coffee_ug", "url": "https://huggingface.co/rgautroncgiar/croppie_coffee_ug" } ], "additional": [ { "label": "Ai Coffee Intelligent Awake (bmz-digital.global)", "url": "https://www.bmz-digital.global/en/ai-coffee-intelligent-awake/" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The Data is fully open-sourced and can be accessed on Hugging Face\n • License: CC BY SA 4.0\n • Data Type: Tabular\n • Key Variables: Coffee cherry color (5), coffee cherry size\n • Appr. No of observations: 8333\n • Format: dbf, csv\n \n The data was collected over the span of 1.5 years covering different crop cycles between 2022 and 2024. The Data Collection was conducted by Alliance Bioversity & CIAT and M-Omolumisa via on-the-ground visits of the farms.\n\nFor this dataset / use case, a responsible AI Assessment was undertaken to help AI developers and project managers to identify, assess and mitigate potential harms and biases in AI. For methodology, see https://www.bmz-digital.global/en/news/ethical-crash-test-for-ai-how-to-navigate-the-road-to-responsible-innovation/", "provenance": "curated" }, "model_characteristics": { "text": "Processed Data:\n The dataset is made of a mix of Arabica and Robusta coffee tree parts (with and without a background isolation element) with individual bounding boxes around all coffee cherries. These RGB pictures were on-farm collected with smartphones with the collaboration of smallholder farmers. \n Code for Crop Classification Model:\n Code to create the training data and for training the models for crop mapping can be accessed here: \n - Dataset\n - Model\n Application:\n The dataset/model can be used for automated cherry count or coffee ripeness assessment.", "provenance": "curated" }, "how_to_use": { "text": "This resource can be of use to anyone interested to automate the task of manual cherry counts (the traditional approach to harvest estimation) and to estimate coffee harvest on a plot level. This is done by processing pictures through AI-supported image recognition software along with a manual branch count (= tree level estimation) and a tree-per-ha estimation (or recall data) (scaling to plot level). Thus, this system is useful for anyone interested in building a central system for managing yield data at the farmer level. It also provides the option to share this data with external users; with for example cooperatives having the possibility to feed information from the application in an integrated database of their member’s performance and farm characteristics, provided they obtained prior consent to do so. This would then scale the Croppie system to have an aggregated data view of multiple users in a region, e.g. as a cooperative level dashboard.\n\nThis use case also included the development of a business model and funding model for open source AI as a stepping stone towards financially viable operations. It was facilitated by Villgro Africa's and FAIR Forward's six-month mentorship programme on \"Creating sustainable business and funding models with open source AI\". See: https://www.bmz-digital.global/en/news/open-source-ai-business-impact/", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_78/images/croppie.png" }, { "id": "ui_79", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_79-finding_good_spots_for_decentralized_green/", "aliases": [ "bridging_the_energy_gap_machine_learning", "finding_good_spots_for_decentralized_green", "scaling_mlbased_site_identification_for_minigrids" ], "title": "Finding good spots for decentralized green energy grids in Uganda - AI based site identification for MiniGrids", "description": { "text": "The Site Identification tool is an AI-driven tool to enhance renewable energy planning. This tool utilizes machine learning and satellite imagery to identify optimal sites for renewable energy deployment, particularly in remote regions. A successful pilot in Lamwo District demonstrated its potential to improve energy access by recommending suitable electrification strategies for villages. This initiative aligns with Uganda’s National Electrification Strategy, which seeks to connect 10 million households to the most optimal energy source by 2030.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Uganda", "iso2": "UG" } ], "regions": [], "sdgs": [ "SDG 13" ], "data_types": [ "Geospatial/Remote Sensing", "Tabular" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Sunbird.ai", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Sunbird.ai (emwebaze@sunbird.ai)", "access_note": null, "links": { "dataset": [ { "label": "GitHub: lamwo-electrification-project", "url": "https://github.com/SunbirdAI/lamwo-electrification-project/tree/main" } ], "usecase": [ { "label": "lamwo.sunbird.ai", "url": "https://lamwo.sunbird.ai/" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "-• Data Type: Geospatial and tabular \n• Key Variables: Solar Irradiance,Biomass Resources,Wind Speed and Consistency,Biomass Resources,Hydropower Potential,\nPopulation Density,Proximity to Existing Grids,Road Networks,Land Availability,Protected Areas,Cost of Energy Delivery\n• The data adheres to FAIR (Findable, Accessible, Interoperable, Reusable) principles and includes anonymization measures to protect sensitive information.\n\nFor this dataset / use case, a responsible AI Assessment was undertaken to help AI developers and project managers to identify, assess and mitigate potential harms and biases in AI. For methodology, see https://www.bmz-digital.global/en/news/ethical-crash-test-for-ai-how-to-navigate-the-road-to-responsible-innovation/", "provenance": "curated" }, "model_characteristics": { "text": "This web application is a decision support tool. It is built as a prototype for a large-scale decision support tool for determining electrification strategies for different geographical areas. Currently implemented in Lamwo district electrification.", "provenance": "curated" }, "how_to_use": { "text": "This web application is a prototype decision support tool designed to guide electrification strategies across various geographic areas. It offers an initial village-level assessment through a color-coded map, with the legend explaining each category. Users can explore detailed analyses at the village level—the smallest unit for assessing electrification needs. The tool evaluates both demand and supply factors and suggests the most suitable energy solutions, including biomass, wind, solar home systems, mini-grids, or grid extension. It is intended to support government officials, energy sector stakeholders, and development partners in making informed renewable energy decisions.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_79/images/solar_grid.jpg" }, { "id": "ui_80", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_80-estimating_solar_irradiance_for_improved_solar/", "aliases": [ "development_of_unbiased_ai_models_for", "estimating_solar_irradiance_for_improved_solar" ], "title": "Estimating Solar Irradiance for Improved Solar Energy Planning in Sub-Saharan Africa through AI", "description": { "text": "The project successfully developed a machine learning model to predict daily Global Horizontal Irradiance (GHI) in Sub-Saharan Africa, with a specific focus on Uganda. The core objective was to correct the systematic bias found in readily available satellite-derived solar irradiation data (like NASA CERES and CAMS). Key activities included identifying five installation sites in Uganda for ground-truth data collection (Gulu, Mbale, Jinja, Kasese, Hoima), procuring pyranometers (though installation was hindered by lack of Ministry of Energy approval), and extensively collecting data from four satellite sources and partner ground stations (Cross Boundary Energy, Makerere University, Ministry of Energy) combined with local topological information (ie. altitude). A Random Forest model was trained on this integrated dataset, proving superior to other models (LSTMs, SVMs) by effectively adjusting the satellite predictions to align with the ground truth across all climatic regions, thus providing a significantly more reliable and bankable estimate of solar resources essential for rural electrification planning.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Uganda", "iso2": "UG" } ], "regions": [], "sdgs": [ "SDG 13" ], "data_types": [ "Geospatial/Remote Sensing", "Tabular", "Meterological" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Makerere University (Marconi Lab)", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Jan Mikelson Andrew Katumba katraxuk@gmail.com", "access_note": null, "links": { "dataset": [ { "label": "GitHub: Solar_irradiation", "url": "https://github.com/Marconi-Lab/Solar_irradiation" } ], "usecase": [ { "label": "irradiation-portal-55883164704.europe-west1.run.app", "url": "https://irradiation-portal-55883164704.europe-west1.run.app/" }, { "label": "GitHub: Irradiation_Portal", "url": "https://github.com/Marconi-Lab/Irradiation_Portal" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "Bias-corrected daily Global Horizontal Irradiance (GHI) data for Uganda, derived by correcting CAMS and NASA POWER satellite datasets against ground-truth measurements from 56 validation sites across Uganda. Training data includes measurements from 7 African countries. The correction addresses a 20% satellite overestimation of solar potential. Data is accessible via the Solar Irradiance Portal web interface and a RESTful API for application integration.\n\nSource: https://github.com/Marconi-Lab/Solar_irradiation, https://irradiation-portal-55883164704.europe-west1.run.app/", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Random Forest model (SuSSE -- Sub-Saharan Solar Estimation) that corrects satellite-derived solar irradiance estimates for Uganda. Input: CAMS and NASA POWER satellite GHI data. Output: bias-corrected daily GHI values. Validated against 56 ground-truth sites across Uganda, achieving R-squared of 0.86 and correcting a 20% overestimation in standard satellite products. Application: deployed as an interactive web portal with maps, charts, and a RESTful API for integration. Code: Python package at github.com/Marconi-Lab/Solar_irradiation (87.9% Jupyter Notebook, 12.1% Python).\n\nSource: https://github.com/Marconi-Lab/Solar_irradiation, https://irradiation-portal-55883164704.europe-west1.run.app/", "provenance": "auto-enriched" }, "how_to_use": { "text": "## How to Use This Resource\n\nThis resource addresses a critical problem for solar energy planning in Uganda: standard satellite data sources such as CAMS and NASA POWER systematically overestimate solar irradiance by roughly 20%, which reduces lifetime energy savings for consumers by 5-20% and creates financial risk through incorrect system sizing. The Irradiation Portal provides corrected Global Horizontal Irradiance (GHI) data by applying a machine learning model trained on ground-truth measurements from 56 validation sites across Uganda and 7 African countries, achieving an R-squared accuracy of 0.86.\n\nAnyone involved in solar project development, rural electrification planning, or energy policy in Uganda can access the corrected data through the [Irradiation Portal web application](https://irradiation-portal-55883164704.europe-west1.run.app/). The portal offers interactive data exploration with dynamic charts and maps, monthly irradiance measurements in kWh/m2/day, and historical data access -- all through a standard web browser. Technical documentation and user support are built into the portal.\n\nFor organisations that want to integrate the corrected solar data into their own planning tools or applications, a RESTful API is available through the portal. This enables automated data retrieval for feasibility studies, system design calculations, or monitoring dashboards.\n\nResearchers and developers looking to extend the underlying methodology can access the SuSSE (Sub-Saharan Solar Estimation) model code, which is openly available on [GitHub](https://github.com/Marconi-Lab/Solar_irradiation). The repository includes the full codebase with Jupyter Notebooks and Python scripts, along with testing and development tools. This opens possibilities for adapting the correction model to other Sub-Saharan African countries where similar satellite overestimation issues exist.\n\nSources:\n- https://github.com/Marconi-Lab/Solar_irradiation\n- https://irradiation-portal-55883164704.europe-west1.run.app/", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_80/images/placeholder_image.jpeg" }, { "id": "ui_81", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_81-ai_as_a_helping_hand_to/", "aliases": [ "ai_as_a_helping_hand_to", "nlp_use_case_strengthening_relevance_of" ], "title": "AI as a helping hand to understand audit reports - A conversational chatbot answering questions related to audit reports of the Auditor General Office of Uganda", "description": { "text": "In Uganda, the Civil Society and Budget Advocacy Group (CSBAG) in partnership with GIZ, has recognized the need for further enhancement in how audit reports are processed and understood. An AI-powered conversational assistant was designed to help anyone understand audit reports. It will answer questions by using Audit reports published by Auditor General Office, Uganda. The prototype tool leverages AI to offer capabilities such as summarizing complex texts, extracting thematic insights, and enabling interactive, user-friendly analysis through a chatbot interface. By making the audit reports more accessible, this aims to increase readership and utilization among stakeholders, which can lead to better accountability and improve service delivery.", "provenance": "curated" }, "kind": [ "usecase" ], "countries": [ { "name": "Uganda", "iso2": "UG" } ], "regions": [], "sdgs": [ "SDG 13" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Civil Society and Budget Advocacy Group (CSBAG)", "links": [] }, "catalyzed_by": { "text": "GIZ Country Office Uganda, FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "CSBAG (csbag@csbag.org)", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "Hugging Face: audit_assistant", "url": "https://huggingface.co/spaces/GIZ/audit_assistant/tree/main" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "Data comes from audit documents in Uganda and samples can be sourced from HuggingFace. Full implementation will come later.", "provenance": "curated" }, "model_characteristics": { "text": "The prototyped is linked - full implementation will come later", "provenance": "curated" }, "how_to_use": null }, "image": "https://fair-forward.github.io/datasets/projects/ui_81/images/placeholder_image.jpeg" }, { "id": "ui_82", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_82-monitoring_deforestation_predicting_landuse_and_la/", "aliases": [ "monitoring_deforestation_predicting_landuse_and_la" ], "title": "Monitoring Deforestation, predicting Landuse and Landcover changes and planning forest restoration in Uganda through AI-powered Remote Sensing", "description": { "text": "This project provides Uganda’s first openly accessible AI-ready satellite imagery dataset designed to predict land-use and land-cover change. It was created to address persistent deforestation and the lack of reliable, context-specific data needed to detect, forecast, and manage ecosystem degradation. Through an open-source replication kit including annotated Sentinel-2 and Hansen datasets, protocols, and inventory data, this products enables local innovators, forest agencies, and policymakers to build AI tools for monitoring forests, planning restoration, and supporting NDC and SDG15 actions. Its goal is to improve early detection of land-cover change, strengthen climate-smart decision-making, and empower local institutions with high-quality geospatial data.", "provenance": "curated" }, "kind": [ "usecase" ], "countries": [ { "name": "Uganda", "iso2": "UG" } ], "regions": [], "sdgs": [ "SDG 13", "SDG 15" ], "data_types": [ "Geospatial/Remote Sensing", "Images" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": null, "organizations": { "provided_by": { "text": "Makerere University in cooperation with the National Forestry Authority (NFA) of Uganda", "links": [] }, "catalyzed_by": { "text": "Lacuna-Fund / Meridian (Climate-call) & FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Makerere AI Lab - Dr. Joyce Nabende (joyce.nabende@mak.ac.ug), National Forestry Authority (NFA): https://nfa.go.ug/", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "1drv.ms: Es68F3WluONJjORAR_WvTloBdtisO9TGs…", "url": "https://1drv.ms/f/c/BFFEF9896BFF8791/Es68F3WluONJjORAR_WvTloBdtisO9TGsGyWR8Se5xNYDA?e=q0Afso" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "1. Landus Landcover Datasets (2015 -2023), \n2. ArcGIS Pro Project, ArcGIS DEEPLearning Model for Tree Species.\n\nThe dataset provides georeferenced, annotated Sentinel-2 and Hansen imagery with clear landuse and landcover classes, groundtruth data, and standardized protocols. Users can train AI models for landuse and landcover classification, detect deforestation, validate predictions, and support restoration planning. Its high resolution, temporal coverage, and open accessibility enable accurate, scalable environmental monitoring.", "provenance": "curated" }, "model_characteristics": { "text": "The AI application analyzes historical satellite imagery gereated landuse and landcover maps (2015-2024) and predicts future landuse and land-cover change, offering simple inputs including Sentinel-2 and Hansen datasets of tree cover and outputs such as classified maps and simulations for 2025 and 2030. It identifies deforestation, farmland expansion, and urban growth patterns, supporting evidence-based land management. While highly accurate (ANN: 85.2% OA), limitations include data inconsistencies, sampling bias, and reliance on remote-sensing quality. Ethical use requires adherence to responsible-AI assessment, transparency, and proper citation. The model runs on standard GIS/AI software (QGIS, google colab, Jupyter-notebook) with no special hardware. All derived datasets and protocols are open for reuse under an open license; users must credit the original project to strengthen community knowledge and track downstream impact.", "provenance": "curated" }, "how_to_use": { "text": "One can immediately use this AI system to classify landuse and landcover types, detect deforestation, and generate LULC predictions for 2025–2030. Users can replicate the full workflow from preprocessing imagery, training ANN models, and simulating future change scenarios to support forest monitoring, restoration planning, district land-use decision-making, or environmental reporting. Researchers can extend the work by integrating socio-economic variables, testing hybrid models, adding additional years of imagery, or adapting the model to new districts. Replication requires awareness of limitations such as sampling imbalance, image quality differences, and potential classification bias; an ethical-AI review is recommended for all downstream use. Costs remain low because the tools (QGIS, GEE, SEPAL) and datasets are free; only compute for model training (standard PCs) and technical time are required. All datasets, protocols, and documentation form an open public-good resource, with opportunities for collaboration through Makerere AI Lab and NFA.", "provenance": "curated" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_82/images/placeholder_image.jpeg" }, { "id": "ui_85", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_85-enhancing_business_registration_in_kenya_through/", "aliases": [ "brschatbot", "enhancing_business_registration_in_kenya_through" ], "title": "Enhancing business registration in Kenya through a chatbot", "description": { "text": "The BRS-chatbot is an AI-powered chatbot that streamlines the business registration process in Kenya. It aimes to enhance access to information, simplify the registration process, improve user experience through Human-Centered Design principles, and promote compliance with post-registration requirements. The goal is to reduce reliance on intermediaries and make the registration process more user-friendly and efficient", "provenance": "curated" }, "kind": [ "usecase" ], "countries": [ { "name": "Kenya", "iso2": "KE" } ], "regions": [], "sdgs": [ "SDG 8" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case", "tags": [ "dataset", "model", "pilot", "usecase" ] }, "license": null, "organizations": { "provided_by": { "text": "Made by People / Tanasuk Technologies Africa, Kenya, CLEAR Global, Tech Innovators Network Kenya (THiNK), BRS - Business Registration Service (Kenya)", "links": [] }, "catalyzed_by": { "text": "Smart Development Fund, FAIR Forward - AI for All, Digital Transformation Centre Kenya, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Made by People, Kenya (hello@made.ke)", "access_note": null, "links": { "dataset": [], "usecase": [ { "label": "brs.go.ke", "url": "https://brs.go.ke/" } ], "additional": [ { "label": "Giz Kenya Fair Forward Ai Powered Chatbot (web.archive.org)", "url": "https://web.archive.org/web/20251116031001/https://made.ke/case-studies/giz-kenya-fair-forward-ai-powered-chatbot/" } ], "documents": [] }, "content": { "data_characteristics": null, "model_characteristics": null, "how_to_use": null }, "image": "https://fair-forward.github.io/datasets/projects/ui_85/images/placeholder_image.jpeg" }, { "id": "ui_86", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_86-kinycomet_automatic_evaluation_of_machine_translat/", "aliases": [ "kinycomet_automatic_evaluation_of_machine_translat" ], "title": "KinyCOMET: Automatic evaluation of machine translation for Kinyarwanda-English", "description": { "text": "Until now, the lack of automatic evaluation tools made Kinyarwanda-English machine translation development slow and expensive, as it required manual human review. We’ve bridged this gap by creating a new evaluation dataset and model. Both are now open-source to support the research community. Our methodology and results are detailed in our upcoming paper for LREC 2026.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Rwanda", "iso2": "RW" } ], "regions": [], "sdgs": [ "SDG 9", "SDG 10" ], "data_types": [ "Text" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "Apache 2.0", "spdx": "Apache-2.0", "url": null }, "organizations": { "provided_by": { "text": "Mbaza NLP community, German Research Center for Artificial Intelligence (DFKI)", "links": [] }, "catalyzed_by": { "text": "Regional AI Hub Rwanda, FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "Chris Prince Mazimpaka chrismazimpaka7@gmail.com", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: kinycomet_dataset", "url": "https://huggingface.co/datasets/chrismazii/kinycomet_dataset" } ], "usecase": [ { "label": "Hugging Face: kinycomet_unbabel", "url": "https://huggingface.co/chrismazii/kinycomet_unbabel" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "4000 rows of parallel sentences English - Kinyarwanda with annotated translation quality as Direct Assessment", "provenance": "curated" }, "model_characteristics": { "text": "Pretrained Model: KinyCOMET-Unbabel, a translation quality estimation model for Kinyarwanda-English (bidirectional). Fine-tuned on Unbabel/wmt22-comet-da using the COMET framework. Input: source text, machine translation, and human reference translation. Output: quality score from 0 to 1. Achieves 0.75 Pearson correlation with human judgments (vs. 0.30 for BLEU), Spearman 0.59, MAE 0.07. A second variant (KinyCOMET-XLM) based on XLM-RoBERTa-large achieves 0.73 Pearson. Trained on 4,323 human-annotated translation pairs scored by 15 annotators following WMT Direct Assessment standards. Install via pip install unbabel-comet; load with comet.load_from_checkpoint(\"chrismazii/kinycomet_unbabel\"). License: Apache 2.0.\n\nSource: https://huggingface.co/datasets/chrismazii/kinycomet_dataset, https://huggingface.co/chrismazii/kinycomet_unbabel", "provenance": "auto-enriched" }, "how_to_use": null }, "image": "https://fair-forward.github.io/datasets/projects/ui_86/images/kinycomet.png" }, { "id": "ui_87", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_87-tunga_agrichatbot_open_source_suite_/", "aliases": [ "tunga_agrichatbot_open_source_suite_", "tunga_chatbot_open_source_suite_" ], "title": "Tunga Agri-Chatbot Open Source Suite - A full system to build call center agent based voicebots in Kinyarwanda e.g. for Agriculture and other sectors", "description": { "text": "Voicebots acting as call center agents—accessible via telephone and capable of speaking local languages—hold immense potential for development cooperation. They provide a vital link to users in rural areas who may lack internet access, have limited literacy, do not speak English or do not own digital devices. We believe voice-driven interfaces are the most intuitive way to bridge the digital divide for these communities.\n\nAs part of the Tunga project, we have developed such a voicebot. To support further innovation in this space, we are releasing a comprehensive suite of foundational AI models and datasets:\n\nText-to-speech model\nCurrently the highest-quality synthetic voice for Kinyarwanda.\nhttps://huggingface.co/C4IR-RW/kinya-flex-tts\n\nKinyaCOLBERT Free\nThe first Kinyarwanda embedding/information retrieval model, essential for building RAG chatbots such as Tunga.\nhttps://huggingface.co/C4IR-RW/kiny-colbert-free\n\nVoice Dataset\nA dataset for training TTS and Speech-to-Text (STT) models.\nhttps://huggingface.co/datasets/C4IR-RW/kinya-ag-tts\n\nInformation Retrieval Dataset\nThe data used to train the KinyaColBERT model.\nhttps://huggingface.co/datasets/C4IR-RW/kinya-ag-retrieval\n\nKinyaBERT\nA foundational BERT-style model serving as the backbone for various Kinyarwanda NLP tasks.\nhttps://huggingface.co/C4IR-RW/kinyabert \n\nSource Code\nAll relevant code for the models and datasets mentioned above.\nhttps://github.com/c4ir-rw/ac-ai-models", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "Rwanda", "iso2": "RW" } ], "regions": [], "sdgs": [ "SDG 10", "SDG 2" ], "data_types": [ "Text", "Voice" ], "maturity": { "stage": "Dataset > Model > Pilot", "tags": [ "dataset", "model", "pilot" ] }, "license": { "name": "Apache 2.0", "spdx": "Apache-2.0", "url": null }, "organizations": { "provided_by": { "text": "Centre for the Fourth Industrial Revolution Rwanda (C4IR), Kigali Natural Language Processing (KiNLP)", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [] }, "financed_by": { "text": "BMZ", "links": [] } }, "contact": "C4IR Rwanda (info@c4ir.rw)", "access_note": null, "links": { "dataset": [ { "label": "Hugging Face: kinya-ag-tts", "url": "https://huggingface.co/datasets/C4IR-RW/kinya-ag-tts" }, { "label": "Hugging Face: kinya-ag-retrieval", "url": "https://huggingface.co/datasets/C4IR-RW/kinya-ag-retrieval" } ], "usecase": [ { "label": "Hugging Face: kinya-flex-tts", "url": "https://huggingface.co/C4IR-RW/kinya-flex-tts" }, { "label": "Hugging Face: kiny-colbert-free", "url": "https://huggingface.co/C4IR-RW/kiny-colbert-free" }, { "label": "Hugging Face: kinyabert", "url": "https://huggingface.co/C4IR-RW/kinyabert" }, { "label": "GitHub: ac-ai-models", "url": "https://github.com/c4ir-rw/ac-ai-models" } ], "additional": [], "documents": [] }, "content": { "data_characteristics": { "text": "Two Kinyarwanda agricultural datasets. (1) TTS Dataset (kinya-ag-tts): 10,480 audio clips (5,242 female, 5,238 male) of voice actors reading agricultural content. Format: WAV mono at 24 kHz, with TSV metadata mapping clip IDs to text. Total size: 2.63 GB. (2) Retrieval Dataset (kinya-ag-retrieval): 984 agricultural passages in Kinyarwanda paired with 19,537 questions. Structured as query-positive-negative triplets for training retrieval models (1,901,200 training triplets, 19,600 dev, 32,900 test). Format: TSV. Includes morphologically parsed versions of passages and queries. Both datasets created by C4IR Rwanda and KiNLP. License: CC-BY 4.0.\n\nSource: https://huggingface.co/datasets/C4IR-RW/kinya-ag-tts, https://huggingface.co/datasets/C4IR-RW/kinya-ag-retrieval", "provenance": "auto-enriched" }, "model_characteristics": { "text": "Two pretrained models for a Kinyarwanda agricultural IVR system. (1) TTS Model (kinya-flex-tts): Multi-speaker text-to-speech based on MB-iSTFT-VITS2 architecture. Converts Kinyarwanda text to speech at 24 kHz with 3 speaker options (2 female, 1 male). Requires PyTorch; implementation via the DeepKIN-AgAI package. (2) Retrieval Model (kiny-colbert-free): Information retrieval model for RAG, fine-tuned from Davlan/bert-base-multilingual-cased-finetuned-kinyarwanda. 0.2B parameters, F32 safetensors format. Trained on agricultural question-passage pairs. Usable via ragatouille library. Both models by C4IR Rwanda and KiNLP. License: CC-BY 4.0.\n\nSource: https://huggingface.co/C4IR-RW/kinya-flex-tts, https://huggingface.co/C4IR-RW/kiny-colbert-free", "provenance": "auto-enriched" }, "how_to_use": { "text": "## How to Use This Resource\n\nThe Tunga Agri-Chatbot Suite provides the building blocks for a Kinyarwanda-language agricultural advisory system delivered via Interactive Voice Response (IVR). It was built to address capacity gaps in Rwanda's agricultural extension services and is implemented by C4IR Rwanda and KiNLP, supported by GIZ and financed by BMZ.\n\nThis suite is relevant for anyone working on agricultural extension, farmer advisory services, or voice-based information delivery in Rwanda or similar contexts. The core idea is that smallholder farmers can call a phone number and receive spoken agricultural guidance in Kinyarwanda -- covering topics such as pest and disease diagnosis, agro-climatic practices, and MINAGRI support programmes -- without needing internet access or literacy.\n\nThe text-to-speech component (kinya-flex-tts) can generate natural-sounding Kinyarwanda speech with three voice options (two female, one male), outputting audio at broadcast quality (24 kHz). A live demo is available at [huggingface.co/spaces/Professor/c4ir-rw-kinyarwandatts](https://huggingface.co/spaces/Professor/c4ir-rw-kinyarwandatts), where you can hear sample outputs before deciding whether to integrate the model. The underlying TTS training dataset contains 10,482 audio clips recorded by two voice actors, all licensed under CC-BY-4.0.\n\nThe passage retrieval component (kiny-colbert-free) matches farmer questions in Kinyarwanda to relevant agricultural knowledge passages. This is the engine that allows the system to find the right answer from a knowledge base when a farmer asks a question. The accompanying retrieval dataset contains 984 agricultural passages, 19,537 related questions, and nearly 2 million training triplets, also under CC-BY-4.0.\n\nThe suite also includes KinyaBERT, a morphology-aware language model for Kinyarwanda used for passage ranking, available in base (107M parameter) and large (365M parameter) variants. All model code, training scripts, and technical documentation are available in the [DeepKIN-AgAI package on GitHub](https://github.com/c4ir-rw/ac-ai-models/tree/main/DeepKIN-AgAI). All components require attribution to C4IR Rwanda and KiNLP.\n\nDevelopers and researchers can extend this work to other crops, regions, or languages, or integrate these components into existing agricultural advisory platforms. The individual models (TTS, retrieval, language understanding) can also be used independently for other Kinyarwanda language technology applications beyond agriculture. All datasets and models are hosted on Hugging Face under the [C4IR-RW organisation](https://huggingface.co/C4IR-RW).\n\nSources:\n- https://huggingface.co/datasets/C4IR-RW/kinya-ag-tts\n- https://huggingface.co/datasets/C4IR-RW/kinya-ag-retrieval\n- https://huggingface.co/C4IR-RW/kinya-flex-tts\n- https://huggingface.co/C4IR-RW/kiny-colbert-free\n- https://huggingface.co/C4IR-RW/kinyabert\n- https://github.com/c4ir-rw/ac-ai-models", "provenance": "auto-enriched" } }, "image": "https://fair-forward.github.io/datasets/projects/ui_87/images/agribot.png" }, { "id": "ui_89", "canonical_url": "https://fair-forward.github.io/datasets/projects/ui_89-datadriven_decisionmaking_for_farmers_to_increase/", "aliases": [ "datadriven_decisionmaking_for_farmers_to_increase", "digital_green__local_dataset_collection", "geospatial_dataset_of_wheat_and_rice" ], "title": "Data-driven decision-making for farmers to increase climate resilience in India", "description": { "text": "Smallholder farmers are crucial contributors to global food production, and in India often suffer most from poverty and malnutrition. These farmers face challenges such as limited access to modern agriculture, unpredictable weather, and resource constraints. To tackle this issue, Digital Green collected data via surveys, offering insights into farming practices, environmental conditions, and crop yields.\n\nThe project developed a scalable data collection model for JEEViKA’s frontline workers, integrating digital tools and geofencing, and empowered farmers with personalized climate-smart advice through Farmer Scorecards.\n\nThe project has contributed to Digital Greens current FarmerChat.", "provenance": "curated" }, "kind": [ "dataset", "usecase" ], "countries": [ { "name": "India", "iso2": "IN" } ], "regions": [], "sdgs": [ "SDG 2" ], "data_types": [ "Tabular" ], "maturity": { "stage": "Dataset > Model > Pilot > Use-Case > Use-Case with Business Model", "tags": [ "dataset", "model", "pilot", "usecase", "business" ] }, "license": { "name": "CC-BY 4.0", "spdx": "CC-BY-4.0", "url": null }, "organizations": { "provided_by": { "text": "Digital Green", "links": [] }, "catalyzed_by": { "text": "FAIR Forward - AI for All, GIZ", "links": [ { "name": "FAIR Forward - AI for All", "url": "https://www.bmz-digital.global/en/overview-of-initiatives/fair-forward/" } ] }, "financed_by": { "text": "BMZ and Sequoia Climate Fund, CISCO Foundation", "links": [ { "name": "BMZ", "url": "https://www.bmz-digital.global/en/digital-transformation-and-development-cooperation/" } ] } }, "contact": "Digital Green (info@digitalgreentrust.org), Viamo and Partners, Tech for Her, DynAg, Digifarm, Opportunity International", "access_note": null, "links": { "dataset": [ { "label": "GitHub: frame-templates", "url": "https://github.com/digitalgreenorg/frame-templates" } ], "usecase": [ { "label": "GitHub: frame-templates", "url": "https://github.com/digitalgreenorg/frame-templates" } ], "additional": [ { "label": "Giz Ff Dg Learning Report.Pdf (digitalgreen.org)", "url": "https://digitalgreen.org/wp-content/uploads/2024/10/GIZ-FF_DG_Learning-Report.pdf" } ], "documents": [] }, "content": { "data_characteristics": { "text": "The data was collected through a survey conducted across multiple districts in India. It consists of a variety of factors that could potentially impact the yield of rice crops. These factors include things like the type and amount of fertilizers used, the quantity of seedlings planted, methods of preparing the land, different irrigation techniques employed, among other features. The dataset comprises more than 5000 data points, each having more than 40 features.", "provenance": "curated" }, "model_characteristics": { "text": "Multiple models are available based on the shared data. Please refer to the GitHub repository.", "provenance": "curated" }, "how_to_use": { "text": "The possibilities include creating a machine learning solution to predict the crop yield per acre of rice or wheat crops in India. Our goal is to empower these farmers and break the cycle of poverty and malnutrition.\n\nThis use case also included the development of a business model and funding model for open source AI as a stepping stone towards financially viable operations. It was facilitated by Villgro Africa's and FAIR Forward's six-month mentorship programme on \"Creating sustainable business and funding models with open source AI\". See: https://www.bmz-digital.global/en/news/open-source-ai-business-impact/", "provenance": "curated" } }, "image": null } ] }