HuskySafe

HuskySafe is a safety app for UW students that keeps you informed & safe on campus. It shows live alerts on an interactive map, helps you report concerns with AI, and plans routes that avoid danger.

Comment

HuskySafe

Inspiration

Campus safety information is scattered across UW Alerts, social media, and word-of-mouth. Students lack a single place to see real-time incidents, report concerns, and plan safe routes. We built HuskySafe to centralize safety information and make it actionable.

What it does

HuskySafe is a real-time safety app for UW students with three core features:

  1. Live Safety Map - Interactive map showing real-time incidents with severity levels, category filters, and proximity warnings
  2. AI-Assisted Reporting - Chat-based incident reporting powered by Claude AI that guides users through the process
  3. Smart Route Planning - AI-generated safe routes that avoid active incidents, use indoor corridors, and consider historical risk data

The app aggregates data from UW Alerts, Reddit, Twitter, and student reports to create a community-driven safety network.

How we built it

Frontend:

  • React 18 + TypeScript + Vite for fast development
  • shadcn/ui components (Radix UI) + Tailwind CSS for beautiful, accessible UI
  • React Router v6 for navigation
  • React Query for state management
  • Mobile-first responsive design

Backend:

  • Express.js proxy server (port 3001) to handle Claude API calls and avoid CORS
  • Environment-based API key management

AI Integration:

  • Claude API (Anthropic) for:
    • Conversational incident reporting assistance
    • Safe route recommendations based on active incidents and preferences

Data Architecture:

  • In-memory database with TypeScript interfaces
  • Mock data for incidents, users, reports, buildings, and routes
  • Structured data models ready for backend integration

Key Features:

  • Real-time incident tracking with severity levels (1-5)
  • Multi-source attribution (UW Alert, Reddit, Twitter, Student Reports)
  • Timeline tracking for incident updates
  • User preferences and notification settings
  • Anonymous reporting options

Challenges we ran into

  1. CORS issues with Claude API - Solved by creating an Express proxy server to handle API calls server-side
  2. Real-time data aggregation - Designed a flexible data model to handle multiple sources with different formats
  3. AI conversation flow - Iterated on prompts and conversation logic to extract structured incident data from natural language
  4. Mobile UX - Balanced information density with usability on small screens using bottom sheets and compact cards
  5. Safety color system - Created a semantic color system (critical, warning, caution, safe) that works across severity levels and categories

Accomplishments that we're proud of

  • ✅ AI integration - Working conversational interface for incident reporting
  • ✅ Multi-source aggregation - Unified view of incidents from different sources
  • ✅ Safety-first design - Color-coded severity system and clear visual hierarchy
  • ✅ Complete user flow - Onboarding through reporting to route planning
  • ✅ Type-safe architecture - Full TypeScript coverage for maintainability
  • ✅ Mobile-first UI - Responsive design with bottom navigation and sheet modals

What we learned

  • AI prompt engineering - How to structure system prompts for consistent, helpful AI responses
  • Proxy server architecture - Building backend proxies to handle API keys securely and avoid CORS
  • Safety UX patterns - Designing interfaces that convey urgency without causing panic
  • Data modeling - Structuring incident data to support filtering, aggregation, and real-time updates
  • Component composition - Using shadcn/ui for consistent, accessible components

What's next for HuskySafe

Short-term:

  • Real map integration (Google Maps/Mapbox) with actual incident markers
  • Push notifications for nearby incidents based on user preferences
  • UW NetID authentication integration
  • Real-time data sync with UW Alert system

Medium-term:

  • Machine learning to predict high-risk areas based on historical data
  • Indoor navigation for building-to-building routes
  • Integration with UWPD systems for verified incident data
  • Social features: community safety tips and verified user reports

Long-term:

  • Expansion to other universities with customizable campus data
  • Integration with campus security cameras (with privacy controls)
  • Emergency response coordination features
  • Analytics dashboard for campus safety administrators

Technical improvements:

  • Replace in-memory database with PostgreSQL/MongoDB
  • Real-time WebSocket connections for live incident updates
  • Offline mode with cached incident data
  • Performance optimizations for large-scale deployment

Built with ❤️ for the UW community

Share this project:

Updates