Skip to content

Adas4044/puzzled

Repository files navigation

Puzzled

40-70% of medical devices in low and middle-income countries are broken. Not because they're useless, but because technicians lack training and repair guidance in their language.

Puzzled turns any phone camera into an AI-powered repair assistant — making manuals make sense, one step at a time.

How It Works

  1. Select your language — Supports languages via AI translation
  2. Choose a repair tutorial — Browse visual guides for medical equipment
  3. Follow step-by-step instructions — Your camera verifies each step in real-time using AI
  4. Get unstuck instantly — One tap connects you to a remote expert via Zoom

The magic: Our AI learns from just 2-3 images per step (few-shot learning), making it practical to create guides for any device.

What Makes It Different

Dual AI Verification

  • Siamese Neural Network (ResNet18): inference, embeddings, cosine similarity verification
  • Claude Vision API: Natural language feedback on why a step is incorrect
  • CLAHE normalization: Works in poor lighting conditions (critical for field use)

Built For The Real World

  • Few-shot learning: Create new guides with just 2-3 photos per step
  • Multilingual: Instant translation to 100+ languages via MyMemory API + i18next
  • Expert escalation: One-tap Zoom integration with Server-to-Server OAuth
  • Mobile-first: Native camera integration, works on any smartphone

Stack: React 19 • TypeScript • Tailwind • FastAPI • PyTorch (ResNet18) • Claude Vision • Supabase • Zoom API

Quick Start

Frontend:

npm install
npm run dev
# Runs on http://localhost:5173

Backend:

cd backend
python3 -m venv venv
source venv/bin/activate  # macOS/Linux (venv\Scripts\activate on Windows)
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env with your Zoom API credentials from https://marketplace.zoom.us/develop/create

# Start server
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000
# Runs on http://localhost:8000

Test it: Visit http://localhost:5173, select a tutorial, and try the camera verification!

📚 Detailed setup, troubleshooting, and development guides: backend/README.md

API Reference

Endpoint Method Description
/verify/verify-step POST Verify camera image against reference step
/api/help POST Create instant Zoom meeting with expert
/verify/dashboard GET Real-time verification monitoring dashboard
/docs GET Interactive API documentation

Interactive docs: http://localhost:8000/docs

Deployment

Backend (Render)

Build Command: cd backend && pip install -r requirements.txt Start Command: cd backend && uvicorn app.main:app --host 0.0.0.0 --port $PORT

Required Environment Variables:

  • ZOOM_ACCOUNT_ID
  • ZOOM_CLIENT_ID
  • ZOOM_CLIENT_SECRET
  • CORS_ORIGINS
  • ANTHROPIC_API_KEY

For detailed deployment instructions, see backend/README.md

Frontend (Vercel/Render Static Site)

Build Command: npm run build Publish Directory: dist

Environment Variables:

  • VITE_API_BASE_URL=https://your-backend.onrender.com
  • VITE_SUPABASE_URL
  • VITE_SUPABASE_ANON_KEY

Built At TreeHacks 2026

Problem: Medical devices sit broken in clinics across low and middle-income countries while technicians lack training resources in their language.

Solution: AI-powered visual repair guides that work on any phone, in any language, with instant expert backup.

Impact: Instead of donating more devices, we help countries fix the ones they already have.


Tech Stack: React 19 • TypeScript • Tailwind CSS • Vite • FastAPI • PyTorch • ResNet18 • Claude Sonnet 4 • Supabase • Zoom API • i18next • MyMemory Translation API

About

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •