Healthcare Multi-Agent Workflow

An intelligent healthcare support system using LangChain and OpenAI with multi-agent architecture.

Features

• 🛡️ Safety Guardrails - Content filtering and PII protection
• 🎯 Intent Classification - Smart routing to specialized agents
• 🩺 Symptom Checker - Emergency detection with hospital routing
• 💊 Multi-Agent Recommendations:
  • 🌿 Ayurvedic remedies
  • 🧘 Yoga therapy
  • 💡 Wellness guidance
• 🏥 Government Schemes - Health insurance and benefits
• 🧠 Mental Wellness - Support and resources
• 📍 Hospital Locator - Find nearby facilities

Project Structure

.
├── src/
│   ├── __init__.py           # Package initialization
│   ├── config.py             # Configuration management
│   ├── schemas.py            # Data models
│   ├── workflow.py           # Main workflow orchestrator
│   └── chains/
│       ├── __init__.py
│       ├── base_chains.py    # Core chain implementations
│       └── specialized_chains.py  # Domain-specific chains
├── cli.py                    # Interactive CLI interface
├── requirements.txt          # Python dependencies
├── .env.example             # Example environment variables
├── .env                     # Your actual API keys (not in git)
└── README.md               # This file

Setup

1. Install Dependencies

pip install -r requirements.txt

2. Configure API Keys

Create a .env file in the project root:

cp .env.example .env

Edit the .env file and add your API keys:

OPENAI_API_KEY=sk-your-actual-openai-key-here
TAVILY_API_KEY=tvly-your-actual-tavily-key-here

Getting API Keys

• OpenAI API Key: Get from OpenAI Platform
• Tavily API Key: Get free key from Tavily (1000 searches/month free)

3. Run the Application

python cli.py

or

./cli.py

Usage

Interactive CLI

The CLI provides a stateful chat interface with history:

$ python cli.py
🏥 Healthcare Assistant - Initializing...
✓ Ready!

Commands: 'exit' to quit, 'clear' to clear history, 'history' to view

You: I have a backache for 2 days

Commands

• exit - Quit the application
• clear - Clear conversation history
• history - View conversation history

Programmatic Usage

from src import HealthcareConfig, HealthcareWorkflow

# Configuration (loads from .env automatically)
config = HealthcareConfig()

# Initialize workflow
workflow = HealthcareWorkflow(config)

# Process query
result = workflow.run("I have a headache and fever")
print(result)

Workflow Architecture

User Query
    ↓
🛡️ Safety Guardrail Check
    ↓
🎯 Intent Classification
    ↓
🔗 Route to Specialized Agent
    ↓
┌─────────────────────┐
│ Government Schemes  │ → Search & Recommend
│ Mental Wellness     │ → Support + Yoga
│ AYUSH Support       │ → Traditional Medicine
│ Symptom Checker     │ → Assess → Multi-Agent:
│                     │   ├─ Emergency? → Hospital Locator
│                     │   └─ Non-Emergency? → Ayurveda + Yoga + Wellness
│ Hospital Locator    │ → Find Facilities
└─────────────────────┘

Security

• ✅ Never commit your .env file to version control
• ✅ The .env file is already listed in .gitignore
• ✅ Built-in guardrails for PII and harmful content
• ✅ Medical emergencies are not blocked and routed appropriately
• ✅ Keep your API keys secure and don't share them

Development

Adding New Chains

1. Create a new chain class in src/chains/specialized_chains.py
2. Add it to src/chains/__init__.py
3. Initialize in src/workflow.py
4. Add routing logic in the run() method

Verbose Debugging

The CLI runs with verbose logging enabled. You'll see:

• Safety check results
• Intent classification
• Chain execution steps
• Agent invocations
• Search queries and results

License

MIT