🏙️ CivicLens

See the problem. Report it. Watch it get fixed. A civic-tech platform for reporting and resolving public infrastructure issues across India.

---

📌 Overview

CivicLens is a full-stack Progressive Web App (PWA) that empowers citizens to report public infrastructure issues — potholes, broken drains, garbage dumps, damaged streetlights, and more — while enabling government fixer teams to prioritize, assign, and resolve them through a real-time accountability loop.

Built on the MERN stack, CivicLens bridges the gap between citizens and civic authorities by combining community reporting, machine learning-assisted validation, GPS-based deduplication, and a transparent resolution pipeline.

---

✨ Key Features

👤 Citizen-Facing (PWA)

• 📸 Photo-first reporting — Upload issue photos; BLIP-2 auto-generates editable captions
• 🤖 ML Relevance Filter — Only civic-relevant photos pass upload validation
• 📍 GPS geo-tagging — Precise location captured at submission
• 🔁 Raise Count — If an issue already exists within a 50m radius, your report increments its priority counter instead of creating a duplicate
• ✅ Accountability Loop — Issue closure requires an after-photo, verified by ML

🛠️ Government Fixer Dashboard

• 🗺️ Dual-layer Heatmap — Color-coded India map showing:
  • Layer 1: Issue density by region
  • Layer 2: Resolution performance by ward/district
• 📋 Issue Queue — Sorted by raise count, recency, and category
• 📂 After-photo Verification — Required to close any issue

🔒 Trust & Safety

• 🚫 False Positive Management — ML filters at upload + community flagging
• ⚠️ One-Strike Policy — Users confirmed to post false reports are permanently blocked
• 🔐 Role-based Access — Citizen, Fixer, and Admin roles with separate dashboards

---

🛠️ Tech Stack

Layer	Technology
Frontend	React.js (PWA)
Backend	Node.js + Express.js
Database	MongoDB (Mongoose ODM)
ML — Captioning	BLIP-2 (image-to-text)
ML — Classifier	Custom relevance classifier (civic vs non-civic)
Geo-processing	GPS coordinates + Haversine formula (50m dedup radius)
Maps	Leaflet.js / Mapbox
Auth	JWT + bcrypt
Storage	Cloudinary / AWS S3 (issue photos)
Hosting	Vercel (frontend) + Render/Railway (backend)

---

🧠 ML Components

1. Image Relevance Classifier

• Purpose: Filters out non-civic images (selfies, food photos, memes) at upload time
• Approach: Binary classification — civic vs not-civic
• Integration: Called as a pre-upload validation step before any issue is created

2. BLIP-2 Auto-Captioner

• Purpose: Generates a human-readable description of the issue from the uploaded photo
• Output: Editable caption pre-filled in the report form (e.g., "Large pothole on road surface with visible water logging")
• Model: Salesforce BLIP-2 (via HuggingFace Inference API or self-hosted)

3. False Positive Detector (Post-submission)

• Purpose: Flags suspicious or non-genuine reports for admin review
• Triggers: Community flags + ML confidence score below threshold

---

📐 Geo-Deduplication Logic

For every new issue submission:
  1. Extract GPS coordinates (lat, lng)
  2. Query all open issues within 50m radius (Haversine)
  3. If match found:
       → Increment raise_count on existing issue
       → Return "Issue already reported" to user
  4. If no match:
       → Create new issue document in DB

This prevents duplicate entries while surfacing high-priority spots through crowd-validation.

---