Abstract

The Council Query Assistant is a Retrieval-Augmented Generation (RAG) system built to serve Masvingo City Council and its residents. It enables users to ask natural language questions about council operations and receive accurate, document-grounded answers. The system ingests structured civic data—such as by-laws, licensing procedures, water distribution, and public notices—and uses semantic search to retrieve relevant content before generating responses with a language model. A lightweight Flask frontend allows users to interact with the assistant via a clean web interface, making the system accessible to non-technical users and deployable in local government environments.

Methodology

The assistant follows a modular RAG pipeline:

1. Document Loading Civic .txt files are loaded from the data/ directory. These files contain quantized information tailored to Masvingo City Council.

2. Text Chunking Each document is split into semantically coherent chunks to improve retrieval granularity.

3. Embedding and Storage Chunks are embedded using a transformer-based model and stored in a vector database (ChromaDB) for fast similarity search.

4. Similarity Search User queries are matched against stored vectors to retrieve the most relevant chunks.

5. Answer Generation Retrieved context is passed to a language model (e.g. Groq or OpenAI) which generates grounded responses.

6. Flask Frontend A minimal Flask app serves as the user interface. It includes:

• A query input field

• Display of retrieved answers

• Optional debug view for retrieved chunks This frontend simplifies deployment and testing, and can be extended with authentication or mobile responsiveness.

Results

The system successfully answers multi-turn civic queries with high contextual relevance. It supports:

• Licensing workflows based on official council forms

• Bill payment guidance using verified platforms

• Retrieval of ward councillor contacts

• Interpretation of statutory by-laws

• Access to health services and digital records

The Flask frontend enables real-time interaction and makes the assistant usable by council staff and residents. Initial testing shows strong alignment between user questions and retrieved content, with minimal hallucination. The assistant improves civic engagement by making council information searchable, conversational, and accessible.

Demo Screenshots

Welcome Screen

Welcome Screen – Masvingo RAG Assistant

Water Billing Query

Query Example – Water Billing and Payments

System Prompt Handling

System Prompt and Query Response