mwangi-george/kenya-constitution-ai-agent

About AI-powered constitutional assistant that enables users to query the Constitution of Kenya (2010) with precise, context-grounded answers and verifiable citations. # 🇰🇪 Katiba AI — Constitution of Kenya Assistant Katiba AI is an AI-powered assistant that enables users to query **The Constitution of Kenya (2010)** and receive **precise, context-grounded answers with verifiable citations**. The system combines retrieval-augmented generation (RAG), OpenAI models, and a modern React interface to deliver an intuitive, ChatGPT-like experience tailored for legal reference. ## ## Project Structure ``` . ├── agent-backend # FastAPI + Pydantic AI + pgvector (RAG pipeline) └── agent-frontend # React + Vite + Tailwind + shadcn UI (chat interface) ``` --- ## Core Features - **Context-aware Q&A** over the Constitution PDF - **Citation-backed responses** (traceable to source text) - **Streaming responses (SSE)** for real-time interaction - **RAG architecture** (embeddings + vector search) - **Modern UI** - **Dockerized backend** with PostgreSQL + pgvector - Modular, maintainable, OOP-based backend design --- ## Tech Stack ### Backend - FastAPI - Pydantic AI (agent orchestration) - OpenAI (LLM + embeddings) - PostgreSQL + pgvector - SQLAlchemy (async) ### Frontend - React + TypeScript (Vite) - Tailwind CSS - shadcn/ui components - TanStack Query --- ## Getting Started ### 1. Backend Setup ```bash cd agent-backend # create env file cp .env.example .env ``` ### 2. Frontend Setup ```bash cd agent-frontend # create env file cp .env.example .env ``` ### 3 Start all services ```bash docker compose up --build ``` Then index the Constitution: ```bash docker compose exec api python -m app.scripts.index_constitution ``` Backend runs at: ``` localhost ``` Frontend runs at ``` localhost ``` --- ## Usage 1. Open the frontend UI 2. Ask any question about the Constitution 3. Receive: - Real-time streamed answer - Supporting citations from the document --- ## Architecture Overview ``` User → React UI → FastAPI → Retriever (pgvector) ↓ Context Injection ↓ Pydantic AI Agent ↓ Streaming Respons …