wakura-mitchell/Masvingo-Civic-MultiAgent-Assistant

Masvingo Civic Assistant is a multi‑agent RAG system for municipal services in Masvingo, Zimbabwe. Built with Flask, LangGraph, ChromaDB, and Groq, it combines specialized agents, tools and web scraping to deliver accurate, real‑time civic information, improving citizen access and engagement with council services. ## Masvingo Civic Multi‑Agent Assistant ## Overview The Masvingo Civic Assistant is a multi‑agent Retrieval‑Augmented Generation (RAG) system designed to improve citizen engagement with municipal services in Masvingo, Zimbabwe. It combines semantic search, structured knowledge integration, and live web scraping to deliver accurate, context‑aware responses about council services such as billing, licensing, incident reporting, and general inquiries. Built as a production‑ready web application, the assistant demonstrates how modern AI can transform public service delivery by providing citizens with 24/7 access to municipal information through an intuitive interface. ## Features Multi‑Agent Architecture: Specialized agents for billing, licensing, incident reporting, and general queries. Intelligent Query Routing: LangGraph orchestration routes queries to the most appropriate agent. Web‑Enhanced Knowledge Base: Live scraping of official council web pages ensures up‑to‑date information. Hybrid Retrieval: Combines vector search (ChromaDB) with structured data (JSON/SQL). Evaluation Framework: Built‑in metrics for precision, recall, F1 score, relevance, and domain classification accuracy. Safety Measures: Domain filtering, input validation, error handling, and fallback mechanisms to prevent harmful outputs. Production‑Ready Design: Modular architecture, error logging, environment‑specific configuration, and extensibility. ## How It Works Web Scraping & Document Ingestion: Extracts content from official council web pages and local civic documents. Domain Classification: Queries and documents are categorized into domains (billing, licensing, incidents, general, etc.). Vectorization & Storage: Text is embedded using sentence transformers and stored in ChromaDB with metadata. Query Routing: LangGraph orchestrates query flow, directing it to the correct agent. Hybrid Retrieval: Agents pull from both vector database and structured civic records. Response Generation …