Roadmap & Future Features

This page tracks planned enhancements, accepted feature requests, and future capabilities for CivicPulse.

🔜 Planned — Next Sprint

Camera Capture with GPS Injection

Epic: Media
Priority: P1

Allow users to take a photo directly within the app (no gallery needed). GPS coordinates are captured at the exact moment of shutter-press — more accurate than EXIF metadata from the camera hardware.

Approach:

Custom camera UI using navigator.mediaDevices.getUserMedia({ video: { facingMode: 'environment' } })
Live GPS overlay on viewfinder via geolocation.watchPosition()
GPS injected as capturedGeo on the image object at capture time
Falls back to <input type="file" capture="environment"> on older browsers

Content Moderation — Phase 1 (Client-side NSFWJS)

Epic: Safety
Priority: P0

Pre-upload image classification in the browser using TensorFlow.js + NSFWJS model (~7MB). Rejects images classified as Porn (>70%) or Sexy (>85%) before they ever leave the device.

Categories checked: Porn, Sexy, Hentai, Neutral, Drawing

Pros: Zero cost, zero latency, works offline, image never hits server
Cons: 7MB model on first load, can be fooled by edge cases

Content Moderation — Phase 2 (AWS Rekognition)

Epic: Safety
Priority: P1

Backend moderation gate before S3 save. Every uploaded image passes through Rekognition.detectModerationLabels({ MinConfidence: 75 }). Returns structured labels with confidence scores.

Categories blocked: Explicit Nudity, Suggestive (high confidence), Rude Gestures, Drugs
Categories allowed with review: Graphic Violence (valid civic evidence), Visually Disturbing
Cost: ~$0.001 per image

Image Flagging — Phase 2 (Backend Persistence)

Epic: Safety / Moderation
Priority: P1

Persist image flags to DynamoDB. Auto-blur after 2 flags. Admin moderation queue with Approve / Remove / Ban actions.

DynamoDB table: civicpulse-media-flags

PK: imageId
flags: [{ userId, reason, createdAt }]
flagCount: number
status: visible | blurred | removed

Issue Flagging — Phase 2 (Backend Persistence)

Epic: Safety / Moderation
Priority: P1

Persist issue flags to DynamoDB. Auto-hide after 5 flags. Separate from image flagging. Admin queue shows full issue context for review.

Face & Licence Plate Blurring

Epic: Privacy
Priority: P2

Automatically detect and blur faces and licence plates in uploaded civic images before storage.

AWS Rekognition DetectFaces → blur bounding boxes
Rekognition DetectText → detect and blur plate patterns
Applied as a post-upload async Lambda job

🧪 Research Phase

Duplicate Issue Detection (Geo + Image Hash + Text)

Epic: Core
Priority: P2
Status: Parked — design complete, implementation deferred

The Problem

Users frequently report the same physical issue from different angles, with different wording, or at slightly different times. This creates noise in the dataset and dilutes upvote signal.

Multi-Signal Scoring

The system uses three independent signals combined into a weighted duplicate score:

duplicateScore =
  geoProximityScore  × 0.55   // strongest — same physical location
  + imageHashScore   × 0.30   // medium — same photo or same scene
  + titleSimilarity  × 0.15   // weakest — same wording

Score	Action shown to user
≥ 0.85	"This issue has almost certainly been reported — upvote instead?"
0.60–0.85	"Very likely the same issue"
0.35–0.60	"Possibly related"
< 0.35	No duplicate warning shown

Signal 1 — Geo Proximity (Haversine Distance)

When GPS is captured on the new issue, compute Haversine distance against all existing issues of the same category.

function haversineMetres(lat1, lng1, lat2, lng2) {
  const R = 6371000;
  const φ1 = lat1 * Math.PI/180, φ2 = lat2 * Math.PI/180;
  const Δφ = (lat2-lat1) * Math.PI/180;
  const Δλ = (lng2-lng1) * Math.PI/180;
  const a = Math.sin(Δφ/2)**2 + Math.cos(φ1)*Math.cos(φ2)*Math.sin(Δλ/2)**2;
  return R * 2 * Math.atan2(Math.sqrt(a), Math.sqrt(1-a));
}

Configurable radius by category:

Category	Radius
Road (pothole)	50m
Traffic signal	20m
Streetlight / Infrastructure	30m
Waterlogging	300m
Hygiene / Garbage	100m
Healthcare / Systemic	N/A

Signal 2 — Perceptual Image Hash (dHash)

Compute a 64-bit perceptual hash of each uploaded image using the browser Canvas API (~5ms, no library needed). Compare Hamming distance against hashes of existing nearby issues.

Hamming distance 0–5   → Virtually identical (same photo)
Hamming distance 6–10  → Same scene, slight crop/resize
Hamming distance 11–20 → Similar location, possibly same incident
Hamming distance 20+   → Different image

The dHash is a dual-purpose computation — the same hash is used for both duplicate detection and the image blocklist (flagged/removed images are added to the blocklist, future uploads with Hamming < 8 are auto-rejected).

Signal 3 — Token-based Title Similarity (Jaccard)

function jaccardSim(a, b) {
  const ta = new Set(a.toLowerCase().split(/\s+/).filter(w => w.length > 3));
  const tb = new Set(b.toLowerCase().split(/\s+/).filter(w => w.length > 3));
  const intersection = [...ta].filter(w => tb.has(w)).length;
  const union = new Set([...ta, ...tb]).size;
  return union === 0 ? 0 : intersection / union;
}

┌─────────────────────────────────────────────────────┐
│ ⚠️  Similar issues found near you (Road · Mumbai)   │
│─────────────────────────────────────────────────────│
│ ● Pothole causing accidents near Bandra    89 ▲     │
│   48m away · In Progress                            │
│   [📍 48m] [🖼 Image: 94%] [Aa Text: 61%]           │
│   [View Issue]  [↑ Upvote instead]                  │
│─────────────────────────────────────────────────────│
│              [My issue is different →]              │
└─────────────────────────────────────────────────────┘

Implementation Plan

Phase	What	Where
Phase 1	Jaccard title match in NewModal, no backend	Frontend
Phase 2	Geo proximity check when GPS captured	Frontend + existing geo data
Phase 3	dHash on upload + in-memory hash comparison	Frontend
Phase 4	Backend `/api/issues/similar` with TF-IDF	Backend
Phase 5	Semantic embeddings (OpenAI / Bedrock)	Backend

Backend Endpoint (Phase 4)

POST /api/issues/similar
{
  "title": "Deep crater on SV Road",
  "city": "Mumbai",
  "category": "road",
  "geo": { "lat": 19.054, "lng": 72.842 }
}

→ Response:
[
  { "id": "ISS-003", "title": "Pothole causing accidents near Bandra",
    "score": 0.82, "distance": 48, "imageMatch": 0.94 },
  { "id": "ISS-001", "title": "Waterlogging near Andheri Station",
    "score": 0.21, "distance": 4200 }
]

📱 Mobile App — React Native

Epic: Mobile Priority: P1 Status: Planned

Overview

A cross-platform iOS + Android app that brings CivicPulse to mobile-first users. The app shares all business logic and API contracts with the web platform — no separate backend needed.

Tech Stack

Layer	Technology
Framework	React Native + Expo SDK 51
Navigation	Expo Router (file-based, similar to Next.js)
State	Same Zustand/Context pattern as web
API	Shared `/api/*` endpoints (same backend)
Auth	Google Sign-In via `expo-auth-session`
Camera	`expo-camera` — native viewfinder with GPS overlay
Push	`expo-notifications` + existing VAPID backend
Offline	`expo-sqlite` for local issue cache
Maps	`react-native-maps` for issue heatmap
Distribution	Expo Application Services (EAS) → App Store + Play Store

Key Features (Mobile-first)

Native Camera — shoot from within the app, GPS injected at capture time
Push Notifications — native OS notifications for issue updates and community replies
Offline Mode — browse and draft reports without connectivity; auto-sync when back online
Civic Score Widget — iOS/Android home screen widget showing your contribution score
Nearby Issues Map — native map view showing issues clustered near the user
Haptic Feedback — upvote, flag, submit actions use vibration feedback
Biometric Auth — Face ID / Fingerprint login for returning users

Project Structure

mobile/
├── app/                  # Expo Router file-based routes
│   ├── (tabs)/
│   │   ├── index.tsx     # Dashboard (same data as web)
│   │   ├── issues.tsx    # Issues list with Near Me sort
│   │   ├── report.tsx    # Report new issue (camera + form)
│   │   └── profile.tsx   # Civic score + settings
│   ├── issue/[id].tsx    # Issue detail with chat
│   └── _layout.tsx       # Root layout + auth guard
├── components/           # Shared UI components
├── hooks/                # Shared hooks (useUserLocation, useFlags, etc.)
├── api/                  # API client (same endpoints as web)
└── app.json              # Expo config

Shared Code Strategy

Business logic in frontend/src/lib/ (pure JS/TS, no React imports) can be consumed by both web and mobile:

haversineMetres() — already portable
fmtDistance() — already portable
API client functions
Scoring algorithm

Implementation Phases

Phase	Milestone
Phase 1	Expo scaffold + auth + issue list + issue detail (read-only)
Phase 2	Report new issue (form + camera + GPS + upload)
Phase 3	Push notifications + offline mode
Phase 4	Heatmap + nearby issues map
Phase 5	Civic score widget + biometric auth
Phase 6	App Store + Play Store submission

📲 Growth & Marketing

Epic: Growth Priority: P1 Status: TODO

Auto-generate shareable issue cards and viral mechanics so every report becomes a distribution channel.

Features:

Open Graph images: auto-generated per issue using Satori — title + category icon + upvote count + city + status badge. No headless Chrome.
One-click share: X/Twitter card, WhatsApp deep link, Instagram Stories image
Dynamic routes: GET /api/issues/:id/og-image returns PNG (cached 10min)
UTM tracking: every shared URL carries ?utm_source=civicpulse&utm_medium=share&utm_campaign=issue for campaign attribution

Backend: backend/services/shareService.js — Satori rendering + S3 cache Frontend: share sheet component with platform icons

Marketing Integrations

Epic: Growth Priority: P2 Status: TODO

Instrument the funnel from landing → sign-in → first report → returning user.

Tool	Purpose	Where
Mailchimp / Loops.so	Waitlist email capture on landing page	Landing CTA
PostHog	Product analytics alongside GA4 — funnel, retention, session recording	`frontend/src/analytics.js`
Hotjar	Heatmaps on issue list + landing page	Loaded async
UTM params	Track campaign effectiveness for every inbound link	`App.tsx` on mount

Referral & Viral Loop

Epic: Growth Priority: P2 Status: TODO

Turn every reporter into a distribution channel.

Refer-a-friend link: unique ref code tied to civic score
City leaderboard badges: shareable PNG showing your rank — "#3 in Mumbai 🏙️"
Social proof on issue detail: "47 people in Mumbai upvoted this"
WhatsApp forward: pre-formatted message with issue URL + summary

Depends on: Social Media Validation Kit (row 65)

💡 Backlog

Feature	Epic	Priority	Notes
PhotoDNA / NCMEC hash matching	Safety	P0 (compliance)	Required if platform scales, for CSAM detection
Bulk issue import from CSV	Admin	P2	For seeding city-specific data
Issue status API webhooks	Core	P2	Notify municipal systems on status change
Multi-language support (Hindi, Marathi, Tamil)	UX	P2	i18n with react-intl
Offline support (PWA)	UX	P2	Service worker caching for offline browsing
Municipal authority dashboard	Admin	P1	Separate portal for government stakeholders
Issue assignment workflow	Core	P1	Assign to specific dept, track SLA
SMS notifications	Notifications	P2	Twilio for non-smartphone users
Heatmap view	Discovery	P2	City-level issue density map
Civic score API (public)	Scoring	P2	Open API for third-party civic apps
Mobile App (React Native + Expo)	Mobile	P1	See Mobile App section above for full spec

🔜 Planned — Next Sprint​

Camera Capture with GPS Injection​

Content Moderation — Phase 1 (Client-side NSFWJS)​

Content Moderation — Phase 2 (AWS Rekognition)​

Image Flagging — Phase 2 (Backend Persistence)​

Issue Flagging — Phase 2 (Backend Persistence)​

Face & Licence Plate Blurring​

🧪 Research Phase​

Duplicate Issue Detection (Geo + Image Hash + Text)​

The Problem​

Multi-Signal Scoring​

Signal 1 — Geo Proximity (Haversine Distance)​

Signal 2 — Perceptual Image Hash (dHash)​

Signal 3 — Token-based Title Similarity (Jaccard)​

UX — Duplicate Panel in New Issue Modal​

Implementation Plan​

Backend Endpoint (Phase 4)​

📱 Mobile App — React Native​

Overview​

Tech Stack​

Key Features (Mobile-first)​

Project Structure​

Shared Code Strategy​

Implementation Phases​

📲 Growth & Marketing​

Social Media Validation Kit​

Marketing Integrations​

Referral & Viral Loop​

💡 Backlog​