OkelloAndrewPeters/eprc-ai-assistant

AI-powered knowledge portal and research assistant for EPRC Uganda # EPRC AI Research Assistant An AI-powered knowledge portal and research assistant for the **Economic Policy Research Centre (EPRC) of Uganda** — built to make 30+ years of Uganda's economic policy research instantly accessible to researchers, policymakers, and the public. > Built entirely in-house by EPRC ICT staff as a zero-cost alternative to external consultants. --- ## What It Does - 🔍 **Semantic search** across 500+ EPRC research documents (1994–2024) - 💬 **Natural language Q&A** — ask questions, get cited answers from real EPRC research - 🌐 **Website navigation assistant** — guides users to the right section of eprcug.org - 📄 **Source citations** — every answer references the exact document and page - 🤖 **Small talk capable** — handles greetings and general queries gracefully --- ## Tech Stack | Component | Technology | |---|---| | LLM | Gemma 4 (via Ollama) — runs 100% locally | | Embeddings | nomic-embed-text (via Ollama) | | Vector Store | ChromaDB (persistent local store) | | Document Registry | PostgreSQL | | Backend API | FastAPI + Python | | Frontend | Vanilla HTML/CSS/JS chat widget | | Document Parsing | PyPDF + python-docx | --- ## Architecture ``` EPRC Documents (PDF/DOCX) ↓ knowledge_base.py — chunks, embeds, stores in ChromaDB ↓ User asks question via chat widget ↓ engine.py — embeds query → searches ChromaDB → retrieves top 8 chunks ↓ Gemma answers using retrieved context + EPRC identity ↓ Streamed response with source citations ``` --- ## Project Structure ``` eprc-ai-assistant/ ├── app.py # FastAPI backend + chat widget UI ├── engine.py # RAG query pipeline ├── knowledge_base.py # Document ingestion pipeline ├── database.py # PostgreSQL + ChromaDB setup ├── prompts.py # System prompts + EPRC identity ├── static/ │ └── index.html # Floating chat widget UI ├── eprc_documents/ # EPRC research documents (not in repo) ├── chroma_store/ # Vector embeddings (not in repo) └── .en …