Build an AI-powered system that can identify, classify, and analyze defects across different industrial domains (manufacturing, textiles, electronics, construction, etc.). The system uses a multi-stage AI pipeline combining Gemini AI for classification and either Hugging Face models or Gemini AI for detailed analysis.

• Uses Gemini 1.5 Flash to classify the defect type
• Identifies which of the 5 categories the defect belongs to
• Returns confidence score, severity level, and initial assessment

• Option A: If Hugging Face API key is available for the category:
  • Uses specialized Hugging Face model for precise detection
  • Returns bounding boxes, confidence scores, and specific defect locations
• Option B: If no Hugging Face API key (fallback):
  • Uses Gemini AI for detailed analysis
  • Provides comprehensive defect description, recommendations, and assessment

1. Surface Scratch Detection

   • Input: Images of metal, wood, or plastic surfaces
   • Detects: Scratches, abrasions, surface damage
   • Use case: Automotive/metal parts inspection
   • HF Model: Optional (facebook/detr-resnet-50)

2. Crack Detection in Concrete/Buildings

   • Input: Drone images of walls, bridges, or roads
   • Detects: Cracks, structural damage, fissures
   • Use case: Civil engineering defect monitoring
   • HF Model: Optional (facebook/detr-resnet-50)

3. Textile/Fabric Defect Detection

   • Input: Fabric roll images
   • Detects: Holes, missing threads, color irregularities
   • Use case: Textile manufacturing QA
   • HF Model: Optional (facebook/detr-resnet-50)

4. PCB (Printed Circuit Board) Defect Detection

   • Input: PCB board images
   • Detects: Soldering issues, missing components, trace damage
   • Use case: Electronics manufacturing QA
   • HF Model: Optional (facebook/detr-resnet-50)

5. Glass/Bottle Defect Detection

   • Input: Industrial glass product images
   • Detects: Cracks, bubbles, surface irregularities
   • Use case: Glass manufacturing QA
   • HF Model: Optional (facebook/detr-resnet-50)

• VITE_GEMINI_API_KEY - Your Google Gemini API key (used for all classifications and fallback analysis)

Add these to .env file for enhanced detection capabilities:

VITE_HF_SURFACE_SCRATCH_KEY=""    # For surface scratch detection
VITE_HF_CRACK_CONCRETE_KEY=""     # For concrete crack detection
VITE_HF_TEXTILE_FABRIC_KEY=""     # For textile defect detection
VITE_HF_PCB_BOARD_KEY=""          # For PCB defect detection
VITE_HF_GLASS_BOTTLE_KEY=""       # For glass defect detection

Note: If Hugging Face keys are not provided, the system automatically falls back to Gemini AI for detailed analysis.

1. User uploads an image through the web interface
2. Gemini AI classifies the defect category automatically
3. System routes the image to the appropriate analysis method:
   • If HF API key exists → Use specialized HF model
   • If no HF API key → Use Gemini for detailed analysis
4. Results are stored in Supabase database with full analysis
5. User views comprehensive report with:
   • Defect type and confidence score
   • Number of defects detected
   • Severity level (low/medium/high)
   • Detailed findings and recommendations
   • Visual annotations (if HF model provides bounding boxes)

• Frontend: React + TypeScript + Vite + Tailwind CSS + shadcn/ui
• Backend: Supabase (Database + Authentication)
• AI Models:
  • Google Gemini 2.5 Flash (Classification + Fallback Analysis)
  • Hugging Face Models (Optional specialized detection)
• Deployment: Ready for Vercel/Netlify

• Flexible AI Pipeline: Works with or without Hugging Face keys
• Multi-domain: Covers 5 different industrial use cases
• Production-ready: Includes authentication, database, and responsive UI
• Showcases integration skills: Combines multiple AI services intelligently
• Scalable architecture: Easy to add new defect categories
• Cost-effective: Falls back to Gemini when specialized models aren't available

1. Clone the repository
2. Install dependencies: npm install or bun install
3. Set up environment variables in .env:
   • Add your VITE_GEMINI_API_KEY (required)
   • Optionally add Hugging Face API keys
4. Run development server: npm run dev
5. Upload images and test the multi-stage detection system!