Eli-Keli/sauti-mwananchi-backend

Sauti ya Mwananchi - multi-agent civic participation chatbot backend for Kenya's elections. # Sauti ya Mwananchi Sauti ya Mwananchi is a multi-agent civic participation chatbot for Kenya's elections. It bridges the gap between "I am registered" and "I know what I am voting for" by giving neutral, verified, and practical guidance to voters in plain language. ## Problem we are solving Kenyan youth registered in record numbers, but many still lack clear, trusted, and neutral information about their rights, voting rules, and what to do on election day. This project helps voters understand the process, find where to vote, and avoid misinformation without pushing any political agenda. ## Agent architecture The system uses Google ADK with a coordinator agent that delegates each message to exactly one specialized agent. Sessions are handled by ADK's in-memory session service and are not persisted. Agents: - Msaidizi: front-door and orchestrator, multilingual, handles greetings and unclear queries - Mwalimu: civic educator, answers only from official documents using RAG - Kiongozi: polling station and registration guidance (voters.iebc.or.ke) - Ukweli: fact-checker for election claims and images, returns one verdict - Mwenza: election day companion with short, step-by-step instructions Tools and services: - Google ADK (LlmAgent + InMemoryRunner) for routing and responses - Gemini (text + vision) as the model backend - Vertex AI Search for legal document retrieval - FastAPI for the backend API How they communicate: - main.py receives the request and calls the ADK runner in adk_app.py - the coordinator agent delegates to a sub-agent using transfer_to_agent - ADK stores events in an in-memory session scoped to session_id ## Run locally Requirements: - Python 3.11+ - Gemini API key - Vertex AI Search datastore (optional but recommended) Setup: ``` python -m venv .venv source .venv/bin/activate pip install -r requirements.txt cp .env.example .env ``` Run: ``` uvicorn main:app --host 0.0.0.0 --port 8080 ``` Health check: ``` curl localhost ` …