Scovia-Barigye/Digital-Extension-Worker

Your virtual extension worker. A local RAG system that "reads" agricultural PDFs to answer farmer queries accurately in English, Luganda, Runyankole, and Swahili. # Digital Extension Worker > **Empowering farmers with fast, localized, and verified agricultural advice through AI.** ## Overview The **Digital Extension Worker** is a local, AI-powered agricultural advisory system built to deliver accurate and accessible farming advice. Designed with a "Uganda-First" approach, it bridges the knowledge gap by using **Retrieval-Augmented Generation (RAG)** to "read" local agricultural PDF manuals and provide hyper-specific, hallucination-free answers to farmer queries. Whether interacting via text or voice, farmers receive actionable advice in local languages, grounded entirely in trusted documentary sources. --- ## Table of Contents - Core Features - Tech Stack - Installation \& Setup - Usage Guide - Roadmap - Contributing --- ## Core Features - **Offline Document Knowledge**: Ingests and memorizes agricultural PDF manuals locally without relying on external cloud databases, utilizing **LlamaIndex** and **ChromaDB**. - **Multimodal Voice & Text Inputs**: Farmers can type their questions or speak directly into their devices. Audio is seamlessly transcribed via the **Sunbird STT API**. - **Verified, Grounded Answers**: Powered by the **Sunbird Sunflower** text generation model. Every answer cites the exact source document name and page number, ensuring complete transparency and verifiability. - **Multilingual Support**: Breaks localization barriers by offering extension advice in **English, Luganda, Runyankole, and Swahili**. - **Responsive & Accessible UI**: Features a gorgeous, lightweight frontend built completely in **Streamlit**, styled with a custom responsive "Ruixen" dark-theme design. --- ## Tech Stack - **Frontend / UI**: Streamlit with custom CSS injection. - **Data Ingestion & RAG**: LlamaIndex core pipeline. - **Vector Database**: ChromaDB (Local persistent storage). - **Embeddings**: HuggingFace (`BAAI/bge-small-en-v1.5`). - **AI Models**: Sunbird AI APIs (Speech-to-Text & Sunflower Qwen). - **Language**: …