About the Project
Inspiration
Citizens frequently report civic issues such as potholes, overflowing garbage, and fallen trees, yet complaints often lack the legal and technical context needed for effective action. Municipal officers must search through scattered regulations, standard operating procedures (SOPs), contractor agreements, and historical precedents before making informed decisions. We wanted to build an AI system that doesn't just file complaints—it prepares evidence-backed, legally grounded cases.
What It Does
ACIRP (Autonomous Civic Incident Resolution Platform) is an AI-powered civic incident resolution platform that transforms a simple citizen report into a structured legal petition.
After classifying an incident, ACIRP retrieves a case-specific municipal knowledge bundle containing relevant laws, municipal acts, engineering standards, contractor SLA clauses, citizen rights, and historical precedents. This large evidence context is then optimized using Paritok, reducing the prompt before it reaches the language model while preserving essential information.
The optimized evidence is used to generate a professional petition with statutory references, supporting authorities in making faster and better-informed decisions.
How We Built It
- Frontend: React
- Backend: FastAPI (Python)
- AI: Google Gemini
- Knowledge Layer: Case-specific municipal knowledge base for potholes, garbage, and fallen trees
- Optimization: Paritok Context Optimizer
- Hosting: Firebase (Frontend) and Render (Backend)
The pipeline is:
- Citizen uploads an image.
- AI classifies the civic issue.
- ACIRP retrieves a rich municipal knowledge bundle containing applicable legislation, SOPs, contractor clauses, and precedents.
- Paritok compresses the retrieved context before LLM reasoning.
- The AI generates a legally grounded petition that can be downloaded or printed.
Challenges We Faced
The biggest challenge was integrating Paritok into the actual execution pipeline instead of treating it as an isolated optimization step. We redesigned the flow so that the retrieved municipal knowledge is optimized before petition generation, ensuring the compressed context is genuinely used during reasoning.
Another challenge was designing a structured municipal knowledge layer that produces different evidence depending on the incident type while keeping the generated petitions relevant and consistent.
What We Learned
This project showed us that high-quality AI applications depend as much on knowledge retrieval and context optimization as on the language model itself. By combining a domain-specific knowledge layer with Paritok's context compression, we learned how to make AI systems more efficient while maintaining meaningful reasoning over large evidence bundles.
What's Next
We plan to expand ACIRP with additional civic issue categories, integrate live municipal regulations and open-government datasets, support multilingual petitions, and connect directly with civic grievance portals for automated submission and tracking.
Log in or sign up for Devpost to join the conversation.