Kare100/agrismart-kenya

# AgriSmart Kenya React Native (Expo) + Supabase mobile app for crop disease diagnosis, market prices, and weather advisories for smallholder farmers in Kenya. Built as a bilingual (English/Swahili), mobile-first tool for farmers who are often on mid-range or older phones and unreliable rural connections, not just a demo for ideal conditions. ## Features - **AI-assisted crop disease diagnosis.** A farmer photographs or picks a leaf photo, selects the crop, and the photo is sent to Google's Gemini vision model (`src/lib/geminiVision.js`) with a prompt written specifically for smallholder Kenyan farming. The response is parsed into a disease name, scientific name, confidence score, plain-language explanation, and a numbered treatment plan (`src/lib/geminiParse.js`). - **Photo optimisation for low-end phones and slow connections.** Before a photo goes anywhere, `src/lib/imageUtils.js` resizes it to a 1024px long edge and re-encodes it as 70%-quality JPEG. Raw camera output on a modern phone can be several megabytes; on a rural or metered connection that turns a single diagnosis into a multi-minute wait or a failed upload. This step keeps both the Gemini call and the Storage upload small and fast, with no visible loss of detail for disease identification, and shows a "Processing photo…" state so the app doesn't feel frozen while it works. This is a standardisation/compression step, not photo enhancement, it won't sharpen a badly blurred or badly lit photo, only make whatever photo is taken quick and reliable to send. - **Disease matching against a structured table.** Gemini returns free text, which on its own can't be reliably queried or joined. `src/lib/diseaseMatch.js` fuzzy-matches that text against the app's own `diseases` table so a diagnosis links to a real `disease_id` where possible, while still keeping Gemini's raw text in `ai_disease_name` if no match is found. - **A real "this isn't a plant" path.** If the photo clearly isn't a leaf, the app skips the Stor …