CivicMind AI

Inspiration

Reporting civic infrastructure issues in India is cumbersome. Citizens spend significant time filing complaints, and responses are often delayed. CivicMind AI aims to turn a 20-minute complaint process into 10 seconds, enabling citizens to report issues efficiently while providing municipal authorities with structured, actionable data.

What it does

CivicMind AI is a multi-agent AI system that: Detects civic issues from uploaded images (Vision Agent). Drafts formal complaint text for the issue (NLP Agent). Maps the issue to the appropriate local municipal authority (Authority Agent). Sends the complaint automatically after user approval (Automation Agent). Tracks the complaint status and logs it in a database (Tracking Agent). The system follows a Human-in-the-Loop (HITL) approach to ensure AI-generated drafts are verified by the user before being sent.

How we built it

Backend: Python + FastAPI + SQLAlchemy Frontend: HTML + CSS + Vanilla JavaScript Vision Agent: YOLOv8 trained on Indian road conditions for potholes, garbage, and street issues NLP Agent: OpenAI GPT-4o-mini for formal complaint drafting Database: SQLite for prototyping, scalable to PostgreSQL Automation: SMTP / email API to send complaint emails Tracking & Validation: Ensures accurate logging of complaint status Architecture: Modular agents collaborate asynchronously using FastAPI BackgroundTasks

Challenges we ran into

Achieving real-time multi-agent orchestration, as backend AI processing can be slow. Displaying live updates in the frontend; initially, only terminal logs were visible. Limited civic issue image datasets; YOLOv8 required India-specific training for potholes, garbage, and street issues. Coordinating 7 team members without a dedicated branch workflow initially presented challenges. Accomplishments that we're proud of Delivered a fully functional multi-agent AI system within hackathon time constraints. Implemented Human-in-the-Loop (HITL) for responsible AI actions. Built a modular architecture that allows easy addition of new agents. Live demo showcases image-to-complaint automation. README and Devpost submission clearly credit each team member’s contributions.

What we learned

Asynchronous tasks are critical for real-time AI pipelines. Modular multi-agent collaboration improves maintainability and scalability. Human-in-the-loop ensures AI outputs are trustworthy and safe. Hackathon teamwork requires clear task allocation and communication, especially with a distributed team.

What's next for CivicMind AI

Deploy the YOLOv8 model on the cloud for faster inference and scalability. Integrate RAG-based municipal authority lookup for accurate complaint routing. Add an image validation agent to filter out non-civic or spam uploads. Build a React/NextJS frontend for a polished user experience. Expand dataset for more civic issues (streetlights, waterlogging, public bins, etc.).piration