raphaelndemo/AgriBrain-Ai

AI-powered agricultural intelligence platform built for smallholder farmers in Kenya. # AgriBrain-Ai **AgriBrain-AI** is an advanced, multimodal artificial intelligence assistant engineered specifically for Kenyan farmers. By combining Large Language Models (Gemini 2.5 Flash), Machine Learning yield projections, and real-time PostGIS spatial data, it provides hyper-local agricultural intelligence to optimize yields, prevent market over-saturation (the "Cobweb Phenomenon"), and route local labor. > **Live Demo:** agribrain-ai-1022818522174.… --- ## Table of Contents - Overview - Key Features - Achitectural Layers - Datasets - Notebook Pipeline - Models & Algorithms - Key Outputs - Setup & Installation - Usage - Project Structure - Key Design Decisions - License --- ## Overview AgriBrain breaks the traditional cycle of guesswork in Kenyan agriculture by delivering **data-driven crop recommendations**. It helps farmers choose high-demand, ecologically viable crops that maximize profitability while supporting national food security. ### Key Features Multimodal AI Diagnostics: Farmers can upload images of diseased crops via web or WhatsApp. The system dynamically compresses and analyzes the images using Gemini 2.5 Flash to provide instant agronomic advice. PostGIS Geospatial Routing: Utilizes advanced SQL spatial queries to instantly connect farmers with agricultural laborers (vibarua / area agents) and agrovets within a 10km radius of their precise GPS coordinates. Predictive ML Yield Engine: A Random Forest Regressor trained on 19 complex macroeconomic and agrochemical features (including CPI, exchange rates, and lagged yields) to calculate precise harvest projections. Cascading Market Arbitrage: Queries live KAMIS market databases to calculate local crop saturation. Features a self-healing geographical fallback (Town ➡️ County ➡️ National) to ensure farmers always receive actionable pricing data. Privacy-First Telemetry: Automated tracking of AI interactions with programmatic phone number masking (e.g., 0712*** …