CIMMYT Co-Lab — GenAI advisory in farmers' languages

The challenge

Smallholder maize and wheat farmers in eastern India lose 15–25% of yield to climate variability, pest pressure and information asymmetry. Existing extension services reach less than one farmer in four, and printed advisories arrive too late to influence in-season decisions. CIMMYT needed a way to deliver agronomy-grade advisories at the speed of WhatsApp, in the dialect a farmer actually speaks.

Our approach

Indev partnered with CIMMYT's data science team to build Co-Lab — a generative AI advisory platform that turns CIMMYT's agronomic research corpus into farmer-grade voice and text answers. A no-code logic builder lets agronomists author advisory rules without writing code; the scheduler pushes time-sensitive nudges (sowing window, fertiliser dose, pest scouting) to enrolled farmers in Hindi, Bhojpuri and Maithili. A peer benchmarking module shows a farmer how their plot is performing against neighbours growing the same variety.

Technology stack

• Generative AI (multilingual LLM)
• Semantic search over agronomic corpus
• Rule-engine / logic builder
• Advisory scheduler (cron + push)
• Voice-first UX (IVR + WhatsApp)
• PostgreSQL + Node.js

Impact in the field

Co-Lab is now CIMMYT's primary digital advisory rail for the eastern Indo-Gangetic Plains. The logic builder has cut the time-to-publish for a new advisory from 2 weeks to under 48 hours, and the GenAI layer means farmers receive answers in the dialect they speak — not the language the research was written in.