YoussefYT12/AgriAfricaAI

Offline Agricultural Intelligence for African Farmers # AgriAfrica AI ### Offline Agricultural Intelligence for African Farmers > Built for the Africa Deep Tech Challenge 2026 — "The Laptop LLM Challenge: Build On-Device AI for the Hardware Africa Actually Has" --- ## Problem Smallholder farmers, agricultural students, and extension workers across Africa often lack reliable, on-demand access to agricultural expertise — especially in areas with poor or no internet connectivity, and on low-end computing hardware. Existing AI assistants require cloud APIs and constant connectivity, making them unusable in exactly the conditions where they are needed most. ## Why On-Device AI Cloud-based LLMs are not a realistic option for most target users: - Unreliable or unavailable internet in rural areas - No budget for recurring API costs - Data privacy concerns for agricultural/financial information - Low-end laptops (4-8GB RAM, no dedicated GPU) are the actual hardware available AgriAfrica AI runs **entirely locally**, after a one-time setup, with no dependency on cloud inference. ## African Context & Target Users - **Primary users:** Smallholder farmers, agricultural students, and agricultural extension workers - **Primary languages:** English and Arabic (with graceful handling of mixed-language queries) - **Core use case:** Ask a practical farming question (crop disease, irrigation, fertilization, pest management) and get a grounded, source-based answer — even completely offline. ## System Architecture ``` User Question | v Embedding Model (multilingual, local) | v FAISS Vector Search Full step-by-step installation instructions are in `docs/SETUP.md`. Quick overview: 1. Install a Windows build of `llama.cpp` (CPU release) 2. Download the Qwen2.5-3B-Instruct GGUF model (Q4_K_M) 3. Set up a Python 3.11 virtual environment and install dependencies (`pip install -r requirements.txt`) 4. Run the data pipeline scripts (or use the pre-built FAISS index included in `data/processed/`) 5. Start `llama-server.exe` locally 6. …