ROBERT-ADDO-ASANTE-DARKO/Areo-AI

AI agent for crop disease diagnosis — RAG pipeline with fine-tuned embedding models --- title: Areo AI emoji: ⚡ colorFrom: pink colorTo: red sdk: streamlit sdk_version: 1.44.0 app_file: app.py pinned: false license: mit short_description: An AI agent for diagnosis diseases in crops --- # 🌿 Areo AI **An AI-powered agricultural assistant for crop disease diagnosis, management advice, and knowledge retrieval.** Areo AI combines computer vision, retrieval-augmented generation (RAG), and domain-fine-tuned embedding models to help farmers and agronomists identify crop diseases, understand their causes, and receive actionable treatment and prevention guidance — with voice output and multilingual support. --- ## Overview Areo AI is a full-stack agricultural intelligence system built on three pillars: 1. **Visual Detection** — A fine-tuned YOLOv8 model identifies crop diseases from uploaded leaf images across 18 disease classes. 2. **Knowledge Retrieval (RAG)** — Disease information is retrieved from a Qdrant vector store using a domain-fine-tuned sentence embedding model, then synthesised by a Groq-hosted LLM into clear, actionable responses. 3. **Dynamic Knowledge Expansion** — The system can autonomously crawl the web (via Crawl4AI + DuckDuckGo) to enrich its knowledge base with up-to-date agronomic content extracted using an LLM schema pipeline. Users interact through a Streamlit interface that supports text chat, image upload, text-to-speech output (Kokoro TTS), and real-time translation. --- ## Architecture ``` User Input (image / text) │ ▼ ┌───────────────────┐ ┌──────────────────────────┐ │ YOLOv8 Detector │────▶│ Disease Label + Confidence│ └───────────────────┘ └──────────┬───────────────┘ │ ┌──────────▼───────────┐ │ Query Construction │ └──────────┬────────────┘ │ ┌───────────────────▼────────────────────┐ │ Qdrant Vector Store │ │ (fine-tuned SentenceTransformer index) │ └───────────────────┬────────────────────┘ │ top-k passages ┌──────────▼───────────┐ │ Groq LLM (RAG) │ │ llama3 / mi …