Swasthagi - AI-Powered Healthcare Assistant

Swasthagi is an intelligent healthcare platform that provides AI-driven medical advice and patient management through multi-agent systems. The project demonstrates two different approaches to building collaborative AI agents for healthcare applications.

Overview

Swasthagi includes two implementations:

1. A2A (Agent-to-Agent) - Main Implementation

The primary implementation using Agent-to-Agent communication patterns. Multiple specialized agents run on different ports and collaborate to provide comprehensive healthcare services.

Key Components:

• MedGemma Advisor: Medical symptom analysis and advice (port 7778)
• Doctor Office Agent: Patient registration and dashboard management (port 7779)
• WhatsApp Orchestrator: User-facing coordinator that orchestrates interactions between agents (port 7777)

Architecture: Agents communicate via HTTP/REST APIs, enabling a distributed microservice-like architecture.

2. Teams - A2A-less Multiagent Implementation

An alternative multiagent approach without explicit Agent-to-Agent communication. Uses a team-based coordination model where agents work together in a unified context.

Key Features:

• Memory management and persistent patient records
• Multi-language support with translation capabilities
• WhatsApp integration for direct user access
• Dashboard for patient record visualization

Features

✨ Medical Consultation

• Symptom analysis using medical AI models
• Evidence-based health recommendations
• Integration with local medical models (MedGemma)

📱 WhatsApp Integration

• Direct patient interaction via WhatsApp
• Natural language health queries
• Persistent conversation history

👥 Patient Management

• Patient registration and profile management
• Medical history tracking
• Dashboard visualization of patient records

🌍 Multi-language Support

• Support for Indian languages and Romanized text
• Automatic translation between languages

💾 Persistent Storage

• SQLite-based patient records
• Patient information and consultation history

Prerequisites

• Python 3.10+
• Ollama with models:
  • functiongemma:270m
  • hf.co/unsloth/medgemma-4b-it-GGUF:Q2_K_XL
• API Keys (depending on implementation):
  • Google Gemini API Key (for Google models)
  • Sarvam AI Key (optional, for language models)
  • WhatsApp Business API credentials

Quick Start

A2A Implementation

1. Setup Environment:

   cd a2a
   python -m venv .venv
   # Windows PowerShell
   .\.venv\Scripts\activate
   # Or Linux/Mac
   source .venv/bin/activate

2. Install Dependencies:

   pip install -r requirements.txt

3. Run Services (in separate terminals):

   Terminal 1 - MedGemma Advisor:

   python medgemma.py

   Runs on http://localhost:7778

   Terminal 2 - Doctor Office:

   python doctor_office_agent.py

   Runs on http://localhost:7779 Dashboard: http://localhost:7779/patientspage

   Terminal 3 - WhatsApp Orchestrator:

   $env:GOOGLE_API_KEY="<your_google_api_key>"
   python whatsapp_agent.py

   Runs on http://localhost:7777

Teams Implementation

1. Setup Environment:

   cd teams
   python -m venv .venv
   .\.venv\Scripts\activate

2. Install Dependencies:

   pip install -r requirements.txt

3. Run Dashboard:

   python dashboard.py

   Access at http://localhost:8000

4. Run Main Agent:

   $env:GOOGLE_API_KEY="<your_google_api_key>"
   python main.py

Project Structure

swasthagi/
├── a2a/                          # Main A2A Implementation
│   ├── doctor_office_agent.py    # Patient registry & dashboard
│   ├── medgemma.py               # Medical advice agent
│   ├── whatsapp_agent.py         # WhatsApp orchestrator
│   ├── patients.json             # Patient data storage
│   ├── index.html                # Web interface
│   └── requirements.txt
│
├── teams/                        # Multiagent Team Implementation
│   ├── main.py                   # Team coordination
│   ├── dashboard.py              # Patient dashboard
│   ├── test.py                   # Testing utilities
│   └── requirements.txt
│
├── templates/                    # HTML templates
│   └── dashboard.html
│
├── tmp/                          # Temporary/data files
│   └── records.csv
│
└── README.md                     # This file

Architecture Comparison

Aspect	A2A	Teams
Agent Communication	HTTP/REST APIs	Direct team coordination
Scalability	Horizontal (add agents as services)	Vertical (single process)
Complexity	Higher (distributed)	Lower (unified)
Deployment	Multiple processes/services	Single process
Memory Sharing	Via API calls	Shared context

Dependencies

Core

• agno - AI agent framework
• fastapi - Web framework
• uvicorn - ASGI server
• ollama - Local LLM support

APIs & LLMs

• google-genai - Google Gemini integration
• openai - OpenAI compatibility
• requests - HTTP client

Optional

• sqlalchemy - Database ORM
• beautifulsoup4 - Web scraping (Teams only)
• googlesearch-python - Google search integration

Configuration

Environment Variables

Set these before running:

# Google API
$env:GOOGLE_API_KEY="your_google_api_key_here"

# Optional: Sarvam AI for language models
$env:SARVAM_KEY="your_sarvam_key_here"

Ollama Models

Ensure Ollama is running and has required models:

ollama run functiongemma:270m
ollama run hf.co/unsloth/medgemma-4b-it-GGUF:Q2_K_XL

Usage

Through WhatsApp

Send health-related queries to the connected WhatsApp number. The system will:

1. Understand your symptoms
2. Provide medical advice
3. Register you as a patient (A2A) or store records (Teams)
4. Maintain conversation history

Through Dashboard

• A2A: Visit http://localhost:7779/patientspage to view registered patients
• Teams: Visit http://localhost:8000 to view patient records and history

Direct API Calls

# A2A - Register a patient
curl -X POST http://localhost:7779/register_patient \
  -H "Content-Type: application/json" \
  -d '{
    "name": "John Doe",
    "illness": "Common Cold",
    "description": "Symptoms include cough and fever"
  }'

Performance & Memory

The system is optimized for:

• Local Inference: Uses lightweight models (MedGemma 4B, FunctionGemma 270M)
• Memory Efficiency: Persistent storage prevents data loss
• Low Latency: WhatsApp API integration for real-time responses

Troubleshooting

Issue: "Ollama connection refused"

Solution: Ensure Ollama is running:

ollama serve

Issue: "API key not found"

Solution: Set environment variables before running agents

Issue: WhatsApp not receiving messages

Solution: Verify WhatsApp Business API credentials and webhook configuration

Future Enhancements

• Multi-language support expansion
• Integration with real medical databases
• Appointment scheduling system
• Prescription management
• Telemedicine features
• Mobile app interface

Contributing

Contributions are welcome! Areas for improvement:

• Medical model accuracy and fine-tuning
• UI/UX improvements
• Additional language support
• Performance optimization
• Test coverage

License

[Add appropriate license here]

Acknowledgments

Built with:

• Agno Framework - AI Agent Framework
• FastAPI - Web Framework
• Ollama - Local LLM Support
• Google Gemini API - Advanced LLM

Support

For issues, questions, or suggestions:

1. Check existing documentation
2. Review sample implementations in code
3. Check logs for detailed error information

---

Swasthagi means "Health" in Sanskrit. This project aims to make quality healthcare advice accessible through AI agents.