Authors: Innocent Nyalala
Affiliation: SAAIL Lab, IIT Madras Zanzibar Campus, Tanzania
Venue: Deep Learning Indaba 2025, Proceedings of Machine Learning Research (PMLR), 2026
East African agriculture supports more than 175 million people but faces mounting challenges from climate change, resource constraints, and information access barriers. Current foundation models fail to address the region's computational limitations, linguistic diversity across 200+ languages, and fundamental differences in knowledge systems. This paper presents CARA-FM, a theoretical framework comprising four pillars: Community-Driven Data Architecture, Indigenous Knowledge Systems, Edge-First Model Design, and Participatory Governance. We propose evaluation metrics spanning technical, agricultural, and cultural dimensions. This framework provides a research agenda for developing agricultural AI systems that operate within severe resource constraints and respect local contexts.
| Pillar | Description |
|---|---|
| Community-Driven Data Architecture | Locally sourced, participatory data collection |
| Indigenous Knowledge Systems | Integration of traditional agricultural knowledge |
| Edge-First Model Design | Models operable on 1-4GB RAM devices |
| Participatory Governance | Community involvement in AI deployment decisions |
cara-fm/
├── framework/ # Framework documentation
├── src/ # Prototype implementations (coming soon)
├── notebooks/
└── README.md
@inproceedings{nyalala2026carafm,
title={Culturally Attuned and Resource-Aware Foundation Models for East African Agriculture: A Theoretical Framework and Research Agenda},
author={Nyalala, Innocent},
booktitle={Deep Learning Indaba},
series={Proceedings of Machine Learning Research},
year={2026}
}SAAIL Lab (Sustainable AI for Agriculture and Intelligent Livelihoods) is based at IIT Madras Zanzibar Campus, Tanzania. We build responsible, locally grounded AI solutions for East Africa and the Global South.