KrushiSaathi
AI agricultural advisory, by voice.
An AI-powered agricultural advisory platform. Extended with soil-testing registration, lab-result tracking, and smart recommendations using a Gemini live-API voice integration for farmers.
KrushiSaathi gives farmers agricultural advice through an AI assistant. The gap it was solving when I joined: advice is only as good as the data behind it, and farmers had no structured way to feed their own soil results into the system.
My contribution added that loop. Farmers register a soil-testing request, the platform tracks the sample through the lab, and results are stored against the farmer's profile once they arrive. The recommendation engine then reads those actual soil values instead of generic regional defaults.
The second piece was voice. Many of the platform's users are more comfortable speaking than typing, so I integrated the Gemini live API for a real-time voice conversation — audio in, streamed audio out — grounded in the same profile and soil data as the text advisor.
The stack is Next.js and TypeScript on Firebase, with LangChain orchestrating the prompt pipeline and Vertex AI backing the models. Working as an open-source contributor meant fitting into existing conventions and shipping each feature as a reviewable, self-contained change.