Designing and Testing Agricultural LLMs

 12 February 2026. Designing and Testing Agricultural LLMs

This webinar brought together two teams building agricultural LLMs through different but complementary approaches—offering a behind-the-scenes look at how AI systems are co-designed, refined, and validated with domain experts, farmers and extensionists. 

• Mamoun Alaoui (ai71) presented the development journey of AgriLLM, a domain-specific agricultural LLM built using retrieval-augmented generation (RAG) and trained on curated agronomic datasets. He walked through the co-design process: from data partnerships and ontology building, to model grounding, prompt evaluation, and iterative testing with agricultural experts across CGIAR and partner institutions. 
• Sulakshana Gupta (Viamo) shared lessons from Ask Viamo Anything (AVA) and Viamo’s locally trained agents—two contrasting advisory approaches now being piloted across multiple countries. She highlighted how Viamo integrates farmer feedback loops, rapid user-testing cycles, and real-time analytics to refine advisory responses and guide model improvements. 
• Discussant:Jawoo Koo, Senior Research Fellow, IFPRI
• Moderator: Eliot Jones-Garcia, Senior Research Analyst, IFPRI; PhD Candidate, Wageningen University

Together, the speakers discussed what effective co-design looks like in practice: aligning technical model development with user needs, incorporating field insights early, and building evaluation frameworks that reflect real advisory contexts.