🛰️ PathFinder

AI-powered disaster response navigation. Upload satellite imagery → detect building damage → generate safe evacuation routes.

PathFinder analyzes post-disaster satellite images using a custom-trained YOLO segmentation model, converts damage detections into geo-referenced danger zones, and computes safe pedestrian/vehicle routes that avoid destroyed areas.

Built for [YHacks 2026] (https://devpost.com/software/pathfinder-7t2r3o?_gl=1*14jzjzt*_gcl_au*MTY2Nzc1NjAwMS4xNzc0OTA2MTUy*_ga*NTE1MzAyNjkuMTc3NDkwNjE1Mg..*_ga_0YHJK3Y10M*czE3NzQ5MjQ4NTAkbzIkZzEkdDE3NzQ5MjQ5MjckajQzJGwwJGgw) by GraphoLogiCode.

---

How It Works

Satellite Image → YOLO Detection → Geo-Referencing → Safe Routing → Rescue Plan

1. Upload a satellite image or select a map region
2. AI Detection — YOLO26m-seg identifies damaged buildings (no-damage, minor, major, destroyed)
3. Geo-Reference — pixel masks → lat/lng GeoJSON danger zones
4. Safe Routing — Valhalla calculates routes avoiding danger zones
5. Rescue Plan — GPT-4o generates severity-specific rescue recommendations

---

Quick Start

# Clone
git clone https://github.com/GraphoLogiCode/PathFinder.git
cd PathFinder

# Backend (Python/FastAPI)
cd backend
pip install -r requirements.txt
cp .env.example .env  # fill in your keys
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

# Frontend (Next.js) — in a new terminal
cd frontend
npm install
npm run dev

Open http://localhost:3000 in your browser.

---

Tech Stack

Layer	Technology
AI Model	YOLO26m-seg trained on xBD dataset (10 disaster types, 4 damage classes)
Backend	Python, FastAPI, Shapely, OpenCV
Frontend	Next.js, React, Mapbox GL JS
Routing	Valhalla (self-hosted)
Database	Supabase (PostgreSQL)
AI Analysis	OpenAI GPT-4o-mini

---

API Endpoints

Method	Path	Description
POST	/detect	Upload satellite image → damage polygon masks
POST	/georef	Convert pixel masks → GeoJSON danger zones
POST	/route	Calculate safe route between two points
POST	/analyze/rescue-plan	Generate AI rescue plan for a severity level
POST	/missions/	Save a mission
GET	/missions/	List all missions

---

Model Performance

Trained on the xBD (xView2) dataset — 10 real-world disasters across 5 countries:

Disaster	Location	Type
Tubbs Fire	Santa Rosa, CA	Wildfire
Hurricane Harvey	Houston, TX	Category 4 Hurricane
Hurricane Florence	Wilmington, NC	Category 1 Hurricane
Hurricane Michael	Panama City, FL	Category 5 Hurricane
Hurricane Matthew	Les Cayes, Haiti	Category 4 Hurricane
Puebla Earthquake	Mexico City, Mexico	7.1 Mw Earthquake
Palu Tsunami	Palu, Indonesia	7.5 Mw + Tsunami
Volcán de Fuego	Guatemala	Volcanic Eruption
Missouri River Flooding	Nebraska/Iowa, USA	River Flooding
Thomas/Woolsey Fire	Ventura, CA	Wildfire

4 damage classes: no-damage, minor-damage, major-damage, destroyed

---

Project Structure

PathFinder/
├── backend/              # FastAPI server
│   ├── app/
│   │   ├── main.py       # App entry point
│   │   ├── config.py     # Environment config
│   │   ├── models.py     # Pydantic schemas
│   │   └── routes/       # API endpoints
│   └── requirements.txt
├── frontend/             # Next.js app
│   └── src/
│       ├── app/          # Pages (mission, etc.)
│       ├── components/   # Map, Sidebar, DangerLayer
│       └── lib/          # API client
├── saferoute/
│   ├── ai/runs/          # Trained model weights & metrics
│   └── data/             # xBD dataset (not committed)
├── demo/
│   ├── images/           # Best-damaged satellite samples
│   ├── satellite_collection.py  # Image catalog & gallery
│   └── test_model.py     # Model validation script

---

Configuration

Create backend/.env:

MODEL_PATH=../saferoute/ai/runs/pathfinder-damage/weights/best.pt
VALHALLA_URL=http://localhost:8002
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_KEY=your-key
OPENAI_API_KEY=sk-your-key

---

Team

GraphoLogiCode — YHacks 2026

---

License

MIT