HAVEN

HAVEN turns crowdsourced photos into live, verified hazard maps and safe routes, guiding citizens to shelters while giving responders a clear, real-time view of disaster impact zones and risks nearby.

Comment

HAVEN: Hazard Alert & Visual Emergency Network

Inspiration

Natural disasters are escalating—wildfires, floods, hurricanes—where minutes mean lives. Current systems depend on slow reports, scattered data, and unverified posts, leaving citizens blind in crises.

We built HAVEN to transform crowdsourced images into actionable, AI-verified intelligence for real-time evacuation support.

Vision

Turn every citizen and camera into part of an intelligent safety network.

What It Does

HAVEN is an AI-powered hazard intelligence and evacuation platform that converts images from citizens, cameras, and sensors into verified incidents on a live map, then routes people to safe zones—fast.

Core Workflow

  1. Citizens upload photos of hazards (fires, floods, earthquakes).
  2. AI classifies the image and estimates severity.
  3. Multiple unique reports verify the incident.
  4. HAVEN alerts nearby users and guides them to safety.

Key Features

  • AI Hazard Classification: Uses Roboflow disaster damage models with transformer/CNN fusion to detect hazard type and severity.
  • Multi-Source Verification: Confirms incidents only when 2+ unique devices report the same hazard in proximity.
  • Impact Radius Mapping: Displays dynamic hazard zones instead of static pins.
  • Safe Zone Navigation: Guides users via Google Maps or in-app routing, factoring accessibility.
  • Minimalist, Stress-Resilient UI: Clean full-screen map with verified hazards, heatmaps, and shelters.
  • Voice Route Dictation (TTS): Reads route summaries aloud for hands-free navigation.
  • Admin Dashboard: Lets responders add/manage safe zones and monitor public reports in real time.

How We Built It

Frontend: Next.js + Tailwind (Google Maps SDK, SWR)
Backend: Flask REST API (http://localhost:5000)
AI Layer: Roboflow Disaster Damage Models via API
Verification Logic: Haversine-based incident grouping and multi-user validation
Routing: Finds nearest active safe zone, returns Google Maps deep link

Architecture Goal: Lightweight, modular, and extensible for rapid prototyping and future scalability with Vertex AI or Firebase.

Challenges

  • Balancing Speed vs. Trust: Designing a verification model that filters noise without delaying urgent reports.
  • Minimalist UX: Creating a calm, clear interface usable under panic conditions.
  • Data Model Design: Achieving production-like logic (Incidents, Reports, SafeZones) within 24 hours.
  • Integrating AI Seamlessly: Making model outputs usable for visualization and routing, not just classification.

Accomplishments

  • End-to-End Pipeline in 24 Hours: From citizen upload → AI detection → verified incident → evacuation route.
  • Crowdsourced Safety Verification: Multi-source confirmation reduces false alarms.
  • Extensible Architecture: Simple Flask backend easily upgradable to full cloud stack.
  • Elegant Design: Usable under stress while conveying complex, real-time information.

Impact

  • Smarter Disaster Awareness: Cuts through social media chaos with structured, verified data.
  • Empowers Citizens: Turns bystanders into active safety contributors.
  • Supports Governments & Responders: Provides fast, spatially aware situational insight.
  • Safety-First AI: Goes beyond detection—guides people out of danger.

What We Learned

  • Structure matters: clear report → incident hierarchy prevents chaos.
  • AI’s power lies in integration, not just accuracy.
  • Minimalism in emergencies saves lives.
  • Even simple infrastructure can tell a powerful real-world story.

What’s Next

  • PostgreSQL + Time Decay: Auto-expire stale incidents.
  • Live Camera + Satellite Feeds: Continuous AI classification from infrastructure sensors.
  • Risk-Aware Routing: Pathfinding that avoids danger zones dynamically.
  • Alerting System: Push/SMS notifications for verified incidents.
  • City Analytics Dashboard: Aggregate data for urban safety planning.

Track Alignment

Google — Build the Next Generation of AI Apps

HAVEN represents the next frontier of Agentic, Real-World AI — systems that not only perceive but reason and act in complex, dynamic environments.

  • AI-Driven Reasoning: HAVEN transforms raw visual data into verified, geospatial intelligence — fusing perception, verification, and decision-making in real time.
  • Dynamic Action Loop: Unlike passive models, HAVEN closes the loop by taking intelligent action—alerting users, guiding evacuations, and continuously refining its map of risk zones.
  • Scalable on Vertex AI / Gemini: Designed for deployment on Google’s AI infrastructure, HAVEN’s modular pipeline (Flask + Roboflow + Next.js) can scale to handle multimodal, city-scale safety networks.
  • Embodied Intelligence: This is AI applied to physical impact—bridging data and human safety in real time, a true “AI that acts with purpose.”

Matt Steele — AI Meets the Physical World

HAVEN bridges human sensing and AI understanding:

  • Uses novel, real-world data collection from citizens and cameras.
  • Processes environmental and spatial data to teach AI about real-world risk zones.
  • Creates a new human-data interface where mobile devices and wearables become part of a spatially aware safety grid.

HAVEN isn’t just a hazard map—it's a prototype for the world’s first living, AI-powered safety network. Every photo and every citizen—working together to save lives.

Share this project:

Updates