EdiGich/farmi-pal-v2

FarmiPal — An AI-powered agricultural assistant for smallholder farmers in East Africa. Features agentic crop disease diagnosis, RAG-powered farming chat, real-time market intelligence, and surplus risk detection to prevent post-harvest losses. Built with Next.js, Django, and AMD ROCm. # FarmiPal > An AI-powered agricultural assistant designed to help farmers diagnose crop diseases, access market insights, detect surplus risks, and get real-time guidance — built with a modern, scalable full-stack architecture. --- ## 📋 Table of Contents - Overview - Core Features - 📸 Image Diagnosis - 💬 Smart Chat - 📊 Market Trends - 🌍 Surplus Insights - Architecture - Tech Stack - Project Structure - Getting Started - API Reference - Development Workflow - GPU Integration (Upcoming) - Pre-GPU Checklist - Roadmap - Contributing --- ## Overview FarmiPal is a farmer-first AI platform built to close the information gap between smallholder farmers and the tools they need to make better decisions. It does this through four tightly integrated features: | Feature | What It Does | AI Method | |---|---|---| | 📸 **Image Diagnosis** | Identifies crop diseases from photos | Pretrained vision model + LLM explanation | | 💬 **Smart Chat** | Answers farming questions conversationally | RAG pipeline + localization | | 📊 **Market Trends** | Shows price trends and interprets them | Simple analytics + LLM narrative | | 🌍 **Surplus Insights** | Predicts regional oversupply risk | Heuristics + weather data + LLM explanation | The system uses **Next.js** as a fast, mobile-first UI layer and **Django** as the core backend for AI orchestration, data persistence, and async job handling. All AI logic is decoupled and GPU-ready. --- ## Core Features --- ## 📸 Feature 1: Image Diagnosis > **"Take a photo. Know what's wrong. Know what to do."** ### What It Does A farmer photographs a crop showing signs of disease or stress. FarmiPal runs the image through a pretrained vision model to classify the condition, then passes the classification result to an LLM to generate a practical, localized explanation with specific action steps. ### Why Two Models? The vision model is fast and accurate at classification but produces only a label (e.g., `"maize_leaf_blight"`). That label alone i …