Team FarmVision

FarmVision AI helps smallholder farmers decide what to do on their field this week, in their own language. The problem: farm advice is often generic, arrives in English, and ignores the weather. A farmer who sees leaf spots doesn't know whether to act today or wait for the rain. What it does: the farmer enters a crop, a location and what they see, and can add a leaf photo. An AI agent built on Meta's Muse model then calls tools: a live weather forecast, the last 7 days of rainfall, and a crop calendar. It returns a 7-day action plan whose advice changes with the evidence, for example postponing field work when heavy rain is forecast. How it stays safe: it never gives pesticide doses or product names, it labels its confidence, and it points farmers to local agriculture experts for treatment decisions. If a model answer fails our safety checks, it shows a conservative observation-only plan instead. Built during the event: the agent and tool loop, the Muse integration, the safety validator, the multi-language plans, and the on-device leaf scanner (trained on Plant Village, [94%] on lab photos and [24%] on real field photos, which we report openly). Tech: Meta Muse, React + Vite, Vercel, TensorFlow.js.