roni-kid/rk-agridig

An offline AI crop disease advisor for smallholder farmers in Ghana — runs entirely on-device without internet, providing disease diagnosis, treatment, and prevention advice for maize, pepper, and tomato. # RK AgriDig **Offline Crop Disease AI for African Farmers** An on-device crop disease diagnostic and advisory system for smallholder farmers in Ghana and West Africa. Runs entirely offline on commodity 8GB laptops — no internet, no API costs, no cloud dependency. ## 🌍 The Problem Smallholder farmers across sub-Saharan Africa lack reliable access to crop disease expertise. When pests or diseases strike, farmers often: - Cannot diagnose the problem without expert knowledge - Lack access to actionable treatment advice - Have no internet to reach cloud-based AI services - Cannot afford subscription costs ($20+/month API fees) **Result:** Entire harvests lost to preventable diseases. ## 🚀 The Solution **RK AgriDig** brings AI-powered crop disease diagnosis directly to farmers' laptops — offline, free, and farmer-friendly. **Features:** - 🧐 **Disease Identification** — "What's affecting my crop?" - 🛠️ **Treatment Advice** — "What can I do to fix it?" - 🛡️ **Prevention Guidance** — "How do I prevent this next season?" - 🌐 **Bilingual** — English + Twi (Ghanaian language) - 📱 **Simple UI** — Gradio web interface, no technical knowledge required - ⚡ **Fast** — Real-time diagnosis on 8GB RAM hardware ## 📊 Dataset Built on **GhanaAgricVQA** — a visual question-answering dataset from Ghana with: - **2,361 Q&A pairs** (train: 2,010 | test: 351) - **787 images** from real Ghanaian farms (RAIL dataset) - **3 crops:** Maize, Pepper, Tomato - **26 disease classes** with expert annotations - **English + Twi** translations for accessibility View dataset on HuggingFace ## 🛠️ Technical Stack | Component | Technology | Notes | |-----------|-----------|-------| | **Model** | Phi-3-mini (3.8B) | Quantized to Q4_K_M GGUF | | **Inference** | llama.cpp + Ollama | CPU-only, memory-mapped loading | | **UI** | Gradio | Simple, farmer-friendly web interface | | **Framework** | Python 3.11+ | Minimal dependencies | | **Target Hardware** | 8GB DDR4 RAM, integrated GPU | Ubuntu 22.04 …