FarmPilot

FarmPilot is an AI farming co-pilot for smallholder farmers in Nigeria. It tackles two problems. First, farmers lack timely, local knowledge: Nigeria has roughly one extension worker for every 10,000 farmers, so most plant, fertilize and sell on guesswork. Second, even when farmers know what to do, they often can't act: a farmer may know the city market pays more for tomatoes but cannot pay ₦150,000 up front for a truck, so they sell cheap or watch produce spoil. Nigeria lost an estimated ₦3.5–5 trillion to post-harvest losses in 2025. FarmPilot follows one farmer through the whole season: - Map the farm: the farmer taps "I'm on my farm now". SAM 3.1 outlines every crop field around them on satellite imagery and measures theirs. They confirm it, or walk the corners with GPS. - Farm report in plain words: soil (ISRIC SoilGrids, with each estimate's range and confidence) and 25 years of NASA rainfall, explained simply by Muse Spark. It is honest about uncertainty, e.g. "test your soil before buying lime". - Season plan: Muse Wonder chooses crops, land per crop, varieties and planting windows for the farmer's soil, rain, land, budget and risk, using live prices found by Muse Spark web search. A deterministic engine computes every quantity and naira (shopping list, costs, bad/expected/good-year margins), and a validator checks land, budget, arithmetic, fertilizer rates and dates before the farmer sees the plan. - Calendar, daily advice and poster: every task is dated; rules check each task against the forecast ("wait to plant", "delay urea, heavy rain coming"), Muse Spark writes the morning briefing, and Muse Image paints a shareable season poster. - When to sell: 24 years of World Food Programme market prices (inflation removed) show how prices usually move after harvest and whether storing pays. - Ask by voice: Muse Voice Transcribe plus a Muse Spark agent that answers from the farmer's own plan, calendar, weather and prices. - Connect: the farmer says "I have 100 crates of tomatoes ready on Friday". Muse Spark creates the listing, SAM 3.1 grades the produce from one photo, and a pooling engine puts nearby farmers on one shared truck, splitting cost by load and distance and taking payment from the sale instead of up front. In our demo, a shared truck cut one farmer's transport share by about half. The same engine books shared storage and equipment. FarmPilot is a working web app (Next.js) built on the Meta Model API. Transporters, neighbours, storage and equipment in Connect are labelled demo listings; payments are shown as a flow, not processed.