FarmiPal — AI Agricultural Assistant

Every harvest season across East Africa, thousands of smallholder farmers watch their crops lose value — not because they grew bad food, but because everyone in their region harvested at the same time, flooded the local market, and prices collapsed. Some of that food rots before it ever reaches someone who needs it. FarmiPal is built to change that. FarmiPal is an AI-powered agricultural assistant with four agentic features designed specifically for smallholder farmers in Kenya and East Africa: Crop Disease Diagnosis — Farmers photograph a sick crop and receive an instant diagnosis powered by a vision model (Qwen2-VL-7B running on AMD MI300X via ROCm), followed by a plain-language explanation and step-by-step action plan in Swahili or English. Smart Farming Chat — A RAG-powered conversational assistant grounded in a curated agricultural knowledge base, including Food and Agriculture Organisation (FAO) guides and Kenya Ministry of Agriculture extension bulletins. Answers are localised to East African conditions. Market Intelligence Agent — An agentic loop that autonomously fetches crop prices, weather forecasts, and searches real-time news to explain why prices are moving — not just what they are. Built on top of free APIs, including Open-Meteo and Google News. Surplus Insights — FarmiPal's most unique feature. An early-warning system that detects regional oversupply risk 3–6 weeks before harvest using CHIRPS satellite rainfall data, NOAA ENSO/ONI index, county-level harvest calendars, and heuristic scoring. Buyers and traders can also identify surplus regions before prices collapse, directly connecting supply to demand. The backend is built with Django and Django REST Framework, the frontend with Next.js and Tailwind CSS, and all vision and language model inference runs on AMD MI300X GPU via ROCm on AMD Developer Cloud. FarmiPal's core infrastructure is live, deployed on Render (backend) and Vercel (frontend), with a Postgres database on Render's free tier.