Community Safety Intelligence Platform
Built for the World Wide Vibes Hackathon 2026 | Public Safety, Emergency Response & City Analytics City of Montgomery, Alabama | March 5–9, 2026
SafeLens Montgomery is a bidirectional community safety intelligence platform purpose-built for the city of Montgomery. It replaces raw incident data and alarming crime maps with responsible, context-rich safety narratives powered by Claude Sonnet 4 and Gemini 2.0 Flash.
Montgomery, Alabama carries an extraordinary history, but today, community trust is under pressure from a public safety system stretched beyond its capacity. The police department is significantly understaffed. As a result, 911 queues are overloaded with non-emergency calls, officers lack the AI force-multiplication needed to intelligently prioritize responses, and residents are left with fragmented, decontextualized raw data that creates fear without understanding.
SafeLens acts as a force multiplier for an understaffed department by bridging the staffing gap with AI. It ingests data from six real-time sources, runs them through an AI pipeline that clusters and validates events across sources, and surfaces human-readable safety narratives. These narratives help officials prioritize their resources where they are needed most. Residents, in turn, can report and confirm incidents directly — staying more aware of what is happening around their city while actively contributing to the picture officials rely on.
- Node.js (v18+)
- A Supabase account and project
- An Anthropic API Key (Claude) and a Google Gemini API Key.
git clone https://github.com/angelatyk/safelens-montgomery.git
cd safelens-montgomery
npm installCreate a .env.local file in the root of the project with your keys:
NEXT_PUBLIC_SUPABASE_URL=your_supabase_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key
SUPABASE_SERVICE_ROLE_KEY=your_service_role_key
ANTHROPIC_API_KEY=your_claude_api_key
GEMINI_API_KEY=your_gemini_api_keynpm run devNavigate to http://localhost:3000 to view the residential dashboard, or http://localhost:3000/demo to log into the official dashboard.
cd workers
pip install -r requirements.txt
python scrape_news.py # Scrape latest news from 6 sources
python nlp_pipeline.py # Classify, cluster, score, and generate narratives- Data Ingestion Pipeline: Python workers collect and score data from 6 sources (including Google News, RSS, Twitter/X, and 311 Reports).
- AI Narrative Generation: Claude Sonnet transforms raw incident data into warm, community-oriented safety updates with neighborhood context and resolution time estimates.
- Multi-Layer News Filtering: Relevance scoring at scraper level (>= 0.15), stricter API threshold (>= 0.3), social media exclusion, and noise keyword filtering.
- Incident Card Feed: Front-end feed displaying AI narratives, official statuses, and live community polling.
- Resident Report Submission: Frictionless, potentially anonymous incident reporting with geolocation.
- Official Narrative Queue: Auth-gated dashboard for city officials offering structured decision support, source breakdown, and live polling results.
- Source Cross-Linking: Junction tables link each narrative to its contributing incidents and news articles via
matched_incident_idsfrom the NLP pipeline.
The application utilizes a dual-model AI architecture:
- Gemini 2.0 Flash: Handles fast, high-volume classification, clustering, and credibility scoring.
- Claude Sonnet 4: Generates quality-sensitive, responsible narratives explicitly prompted to avoid stigmatizing language and to estimate resolution times based on verified data.
The Pipeline Flow: Collect → Classify → Cluster (TF-IDF) → Score (Credibility) → Incident Validation → Narrative Generation → Community Feedback Loop
┌─────────────────────────────────────────────────────────────────────────────────┐
│ DATA SOURCES │
│ AL.com RSS │ WSFA RSS │ Google News │ 311 Reports │ Twitter │
└─────────────────────────────────────────┬───────────────────────────────────────┘
│
▼
┌────────────────────────────────┐
│ Python NLP Workers │
│ Collect → Classify → Cluster │
│ Credibility Score 0–100 │
└───────────────┬────────────────┘
│
┌────────────────┴────────────────┐
│ │
┌───────────▼───────────┐ ┌──────────▼──────────┐
│ Gemini 2.0 Flash │ │ Claude Sonnet 4 │
│ Classify & Cluster │────────▶│ Narrative Writer │
└───────────────────────┘ └──────────┬──────────┘
│
┌──────────────▼──────────────┐
│ Supabase PostgreSQL │
└──────────────┬──────────────┘
│
┌──────────────▼──────────────┐
│ Next.js 14 API │
│ + Supabase Realtime │
└──────────────┬──────────────┘
│
┌───────────────────────────┴───────────────────────────┐
│ │
┌───────────▼───────────┐ ┌───────────▼───────────┐
│ Resident View │ │ Official View │
└───────────────────────┘ └───────────────────────┘
- Frontend / API Layer: Next.js 14 (App Router), TypeScript, Tailwind CSS, shadcn/ui.
- Database & Auth: Supabase (PostgreSQL, Auth, Realtime).
- Data Ingestion/Backend: Python NLP workers deployed via Railway.
- AI Models: Gemini 2.0 Flash & Anthropic Claude Sonnet 4.
- Web Scraping: Bright Data SERP API.
- Hosting: Vercel (Frontend) + Railway (Workers).
neighborhoods— name, geojson polygon, safety_score (0-100)incidents— source (montgomery_311 | nlp_cluster), type, lat/lng, neighborhood_id, raw_datanews_articles— headline, source, url, relevance_score, matched_incident_idsresident_reports— category, description, status, anonymous optionnarratives— AI-generated content, title, status, model_used, source_data_hash
narrative_incidents— links narratives to source incidentsnarrative_news_articles— links narratives to source articles
narrative_feedback— resident votes (accurate / not_relevant)source_reputation— dynamic accuracy scores per news source (feeds back into NLP pipeline)narrative_comments— threaded comments (public + internal/official)narrative_public_updates— immutable history of official updates
| Route | Method | Description |
|---|---|---|
/api/narratives |
GET | List narratives with live junction table counts |
/api/narratives/generate |
POST | Generate AI narratives from all data sources |
/api/narratives/[id]/sources |
GET | Fetch linked incidents, articles, reports |
/api/narratives/[id]/feedback |
GET/POST | Resident voting on narrative accuracy |
/api/narratives/[id]/comments |
GET/POST | Threaded comments (public + official) |
/api/narratives/[id]/public-updates |
GET/POST | Official public-facing updates |
/api/narratives/[id]/verify |
PATCH | Update official status |
/api/news |
GET | Filtered public safety news feed |
/api/reports |
POST | Submit resident incident report |
SafeLens acts as a pilot for a deployable SaaS platform. Future roadmap items include:
- Official City-Wide Alert Broadcasts using Supabase Realtime.
- Interactive Map View with Mapbox heatmaps.
- NWS Weather & Event Calendar integrations.
- Extended Civic Engagement including public infrastructure reporting.
- Angela Kwok (Team Lead) - github.com/angelatyk
- Andres Linero - github.com/andres-linero
- Varoun Bajaj - github.com/varounbajaj-proj