Pill Identification and Drug Safety Chatbot 💊

AI-powered medication analysis system using AWS Rekognition and FDA data

🚀 Key Features

• Pill Imprint Extraction - AWS Rekognition-based text detection from pill images
• FDA Data Integration - Real-time drug information lookup from government APIs
• Medication Safety Chat - OpenAI-powered conversations about drug interactions/usage
• Error Detection - Flag potential incorrect pill identification (LASA risks)

⚙️ Core Components

Backend Services

Endpoint	Method	Description
/health	GET	Service health check
/extract_imprint	POST	Process pill images (JPEG/PNG)
/get_pill_info	POST	Fetch drug details from imprint code
/conversation	POST	AI-powered drug safety Q&A

🛠️ Installation

# Clone repository
git clone https://github.com/yourusername/pill-identification-system.git
cd pill-identification-system

# Install dependencies
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Start Redis
docker run -p 6379:6379 -d redis:alpine

# Run application
python server.py

🔧 Configuration

Create .env file:

AWS_ACCESS_KEY_ID=your_aws_key
AWS_SECRET_ACCESS_KEY=your_aws_secret
REDIS_HOST=localhost
REDIS_PORT=6379
OPENAI_API_KEY=your_openai_key
FDA_BASE_URL=https://api.fda.gov/drug

📚 API Documentation

POST /extract_imprint

Request:

curl -X POST -F "image=@pill.jpg" http://localhost:6969/extract_imprint

Response:

{
  "imprint_number": "AN 715",
  "generic_name": "Lisinopril",
  "summary": "Used for high blood pressure treatment..."
}

POST /conversation

Request:

{
  "imprint_number": "AN 715",
  "generic_name": "Lisinopril",
  "user_query": "What are the side effects?",
  "not_this_pill": false
}

Response:

{
  "explanation": "Common side effects include dizziness...",
  "lasa_warning": ["Hydroxyzine"]
}

🧠 Key Implementation Details

AWS Rekognition Integration

class RekognitionTextExtractor:
    def extract_text(self):
        client = boto3.client('rekognition')
        response = client.detect_text(Image={'Bytes': self.image_bytes})
        return self._process_response(response)

Redis Caching Strategy

# Cache key format: <imprint_number>:<generic_name>
cache_key = f"{imprint_number}:{generic_name}"
redis_client.setex(cache_key, 3600, json.dumps(pill_data))  # 1-hour expiry

FDA Data Processing

class HtmlParser:
    def parse_content(self):
        # Web scraping logic for FDA data
        self.output_name = self.soup.find('span', class_='drug-name').text
        self.output_summary = self.soup.find('div', class_='drug-summary').text

🚨 Error Handling

Error Code	Description	Resolution
400	Missing/invalid file	Check request format
404	No pill data found	Verify imprint code
500	Internal server error	Check server logs

🌐 Deployment

# docker-compose.yml
version: "3.8"
services:
  app:
    build: .
    ports:
      - "6969:6969"
    environment:
      - AWS_ACCESS_KEY_ID
      - AWS_SECRET_ACCESS_KEY
      - OPENAI_API_KEY
    depends_on:
      - redis

  redis:
    image: redis:alpine
    ports:
      - "6379:6379"

📝 License

MIT License - See LICENSE

Note: Requires React frontend implementation for full functionality. Replace placeholder FDA scraping logic with actual API integration.