CivicLens AI Case Study — AI Civic Triage & Caching Pipeline

Overview

CivicLens AI is a civic issue reporting platform developed for hackathons. It uses a React frontend, FastAPI backend, and Google Gemini AI to automatically categorize, evaluate severity, and route infrastructure issues (potholes, sanitation, power outages) to municipal departments.

Live Demo App GitHub Repository

🎯 Problem & Motivation

Municipal complaint systems face backlogs due to manual triage. Citizens submit unstructured complaints without category tags or urgency scores, delaying dispatch teams.

⚙️ Technical Constraints

External LLM API calls add latency and financial cost during traffic spikes. Repeated submissions of similar issues (e.g. power outages in the same area) shouldn't re-trigger full AI parsing.

🏗️ Architecture & Caching Strategy

┌─────────────────┐ ┌───────────────────────┐ ┌──────────────────────┐ │ Citizen │ ────► │ React Dashboard │ ────► │ FastAPI Backend │ │ Complaint Input │ │ (Form & Image Upload) │ │ (Request Pipeline) │ └─────────────────┘ └───────────────────────┘ └──────────┬───────────┘ │ ┌─────────────────┐ ┌───────────────────────┐ ┌──────────┴───────────┐ │ Route to Dept │ ◄──── │ Gemini AI Classifier │ ◄──── │ In-Memory Semantic │ │ & Priority Queue│ │ (Category & Urgency) │ │ Cache Check (Hit/Miss│ └─────────────────┘ └───────────────────────┘ └──────────────────────┘

FastAPI Pipeline: Asynchronous backend router managing complaint submissions and status queries.
Semantic Cache: Designed to intercept recurring issue descriptions, significantly reducing redundant external Gemini API calls.
Gemini AI Classification: Assigns category, severity (Low/Medium/Critical), and department routing tags.

⚖️ Challenges & Trade-offs

Trade-off: Cache lifetime settings must balance latency savings with accuracy—overly aggressive caching risks miscategorizing unique complaints with similar wording.

💡 Lessons Learned

1. Implementing basic caching on AI endpoints is essential for hackathon scalability.
2. Structuring LLM outputs with JSON mode enforces reliable frontend rendering.

📌 Planned Improvements

Planned Map view integration for municipal issue clusters.
Planned Automated status notifications via Webhooks.