🏥 SehatAgent - صحت ایجنٹ

Multi-Agent AI Health System for IDEAX92

"Har Pakistani ki Sehat, AI ki Nigrani Mein" (Every Pakistani's Health, Under AI's Care)

🌟 Overview

SehatAgent is a Multi-Agent AI Health System that provides preventive, explainable, and accessible healthcare guidance for Pakistan. Built for the IDEAX92 Hackathon, it demonstrates how specialized AI agents collaborate to analyze symptoms, assess risks, and generate personalized health recommendations.

Key Features

🩺 Symptom Analysis - Analyzes symptoms in English, Urdu, and Roman Urdu
⚠️ Risk Assessment - Identifies health & nutrition risks (Pakistan-specific)
💊 Preventive Guidance - Simple language recommendations
👨‍⚕️ Healthcare Worker Dashboard - Summarized insights for doctors/LHWs
🗣️ Voice Input - Supports Urdu, Punjabi, and English voice
📴 Offline Mode - Works without internet using cached knowledge

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        User Interface                            │
│              (Web/Mobile - English/Urdu/Punjabi)                 │
└─────────────────────────┬───────────────────────────────────────┘
                          │
┌─────────────────────────▼───────────────────────────────────────┐
│                    FastAPI Backend                               │
│                   (Google Cloud Run)                             │
└─────────────────────────┬───────────────────────────────────────┘
                          │
┌─────────────────────────▼───────────────────────────────────────┐
│                  Agent Orchestrator                              │
├─────────────┬─────────────┬─────────────┬─────────────┬─────────┤
│  Symptom    │    Risk     │   Health    │   Safety    │ Offline │
│  Analyzer   │  Assessor   │   Advisor   │   Guard     │ Helper  │
└─────────────┴──────┬──────┴─────────────┴─────────────┴─────────┘
                     │
        ┌────────────┴────────────┐
        │                         │
┌───────▼───────┐       ┌────────▼────────┐
│   Vertex AI   │       │      FAISS      │
│   (Gemini)    │       │  (Local Index)  │
└───────────────┘       └─────────────────┘
        │
┌───────▼───────┐
│   Cloud SQL   │
│  (PostgreSQL) │
└───────────────┘

Agent Roles

Agent	Role	Responsibility
SymptomAnalyzer	Analysis	Extract symptoms from multilingual input
RiskAssessor	Assessment	Identify health/nutrition risks
HealthAdvisor	Recommendation	Generate preventive guidance
SafetyGuard	Ethics	Ensure safe, ethical responses
OfflineHelper	Fallback	Rule-based guidance when offline

🚀 Quick Start

Prerequisites

Python 3.11+
Docker
GCP Account (provided by hackathon)
gcloud CLI installed

1. Clone & Setup

# Clone repository
git clone <your-repo-url>
cd sehatagent

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Copy environment file
cp .env.example .env
# Edit .env with your credentials

2. Configure Environment

Edit .env with your GCP credentials:

GCP_PROJECT_ID=your-project-id
DB_HOST=your-db-host
DB_NAME=sehatagent
DB_USER=your-db-user
DB_PASSWORD=your-db-password

3. Initialize Database

python scripts/init_db.py

4. Build FAISS Index

python scripts/build_faiss_index.py

5. Run Locally

uvicorn app.main:app --reload --port 8080

Visit: http://localhost:8080

🐳 Docker Deployment

Build Docker Image

docker build -t sehatagent:latest .

Run Locally with Docker

docker run -p 8080:8080 --env-file .env sehatagent:latest

☁️ GCP Cloud Run Deployment

Step 1: Authenticate with GCP

gcloud auth login
gcloud config set project YOUR_PROJECT_ID

Step 2: Enable Required APIs

gcloud services enable \
    run.googleapis.com \
    cloudbuild.googleapis.com \
    artifactregistry.googleapis.com \
    aiplatform.googleapis.com \
    sqladmin.googleapis.com \
    speech.googleapis.com

Step 3: Create Artifact Registry Repository

gcloud artifacts repositories create sehatagent-repo \
    --repository-format=docker \
    --location=us-central1

Step 4: Build and Push Docker Image

# Configure Docker for GCP
gcloud auth configure-docker us-central1-docker.pkg.dev

# Build and tag
docker build -t us-central1-docker.pkg.dev/YOUR_PROJECT_ID/sehatagent-repo/sehatagent:v1 .

# Push to Artifact Registry
docker push us-central1-docker.pkg.dev/YOUR_PROJECT_ID/sehatagent-repo/sehatagent:v1

Step 5: Deploy to Cloud Run

gcloud run deploy sehatagent \
    --image us-central1-docker.pkg.dev/YOUR_PROJECT_ID/sehatagent-repo/sehatagent:v1 \
    --platform managed \
    --region us-central1 \
    --allow-unauthenticated \
    --set-env-vars "GCP_PROJECT_ID=YOUR_PROJECT_ID" \
    --set-env-vars "DB_HOST=/cloudsql/YOUR_INSTANCE_CONNECTION" \
    --set-env-vars "DB_NAME=sehatagent" \
    --set-env-vars "DB_USER=your-user" \
    --set-env-vars "DB_PASSWORD=your-password" \
    --add-cloudsql-instances YOUR_INSTANCE_CONNECTION \
    --memory 2Gi \
    --cpu 2 \
    --min-instances 0 \
    --max-instances 10

📚 API Documentation

Main Endpoints

Endpoint	Method	Description
`/api/v1/health/analyze`	POST	Analyze health query
`/api/v1/health/quick-check`	POST	Quick symptom check
`/api/v1/voice/transcribe`	POST	Voice to text
`/api/v1/voice/analyze`	POST	Voice health analysis
`/api/v1/worker/dashboard`	GET	Healthcare worker insights
`/api/v1/offline/analyze`	POST	Offline mode analysis

Example: Health Analysis

curl -X POST "https://your-service-url/api/v1/health/analyze" \
    -H "Content-Type: application/json" \
    -d '{
        "query": "Mujhe 3 din se bukhar hai aur sir mein dard hai",
        "language": "auto"
    }'

Example Response

{
    "success": true,
    "session_id": "abc123",
    "mode": "full",
    "language": "roman_urdu",
    "symptoms_identified": ["fever", "headache"],
    "risk_level": "MEDIUM",
    "recommendations": [
        "Rest and stay hydrated with water and ORS",
        "Take paracetamol for fever",
        "See doctor if fever persists beyond 3 days"
    ],
    "explanation": {
        "summary": "Based on your fever and headache...",
        "agent_reasoning": [...]
    },
    "disclaimer": "This is not medical advice..."
}

🔒 Degraded Mode

SehatAgent works offline using:

Rule-based symptom matching - Pre-defined patterns
Local FAISS index - Cached embeddings
Offline knowledge base - JSON-based health data

Testing Degraded Mode

curl -X POST "http://localhost:8080/api/v1/offline/analyze" \
    -H "Content-Type: application/json" \
    -d '{"query": "bukhar hai", "language": "roman_urdu"}'

🌍 Supported Languages

Language	Input	Output	Voice
English	✅	✅	✅
Urdu (اردو)	✅	✅	✅
Roman Urdu	✅	✅	-
Punjabi (پنجابی)	-	-	✅

📊 Healthcare Worker Dashboard

Access aggregated insights at /api/v1/worker/dashboard:

Total consultations
Common symptoms distribution
Risk level breakdown
Urgent cases requiring attention
Community health trends

🧪 Testing

# Run tests
pytest tests/ -v

# Test specific module
pytest tests/test_agents.py -v

📁 Project Structure

sehatagent/
├── app/
│   ├── main.py              # FastAPI application
│   ├── config.py            # Configuration
│   ├── agents/              # Multi-agent system
│   │   ├── orchestrator.py  # Agent coordination
│   │   ├── symptom_agent.py # Symptom analysis
│   │   ├── risk_agent.py    # Risk assessment
│   │   ├── recommendation_agent.py
│   │   ├── safety_agent.py  # Ethical guardrails
│   │   └── fallback_agent.py # Offline mode
│   ├── api/                 # API endpoints
│   ├── services/            # External services
│   │   ├── vertex_ai.py     # Gemini integration
│   │   ├── rag_service.py   # FAISS RAG
│   │   ├── language_service.py
│   │   └── speech_service.py
│   └── database/            # PostgreSQL models
├── data/                    # Preloaded data
├── scripts/                 # Utility scripts
├── Dockerfile               # Container config
└── requirements.txt         # Dependencies

🏆 Winning Features

Pakistan-Specific Health Knowledge - Typhoid, Dengue, TB patterns
Multilingual Voice Input - Urdu, Punjabi support
Robust Offline Mode - Works in low-connectivity areas
Explainable AI - Shows reasoning for all decisions
Healthcare Worker Support - Dashboard for LHWs
Privacy-First - No raw health data stored

🤝 Contributing

Built for IDEAX92 Hackathon by BiTech Digital

📄 License

MIT License - See LICENSE for details.

🆘 Support

Emergency (Pakistan): 1122
Edhi Foundation: 115

Built with ❤️ for Pakistan's Healthcare

Name		Name	Last commit message	Last commit date
app		app
data		data
scripts		scripts
tests		tests
.gitignore		.gitignore
Dockerfile		Dockerfile
IMPLEMENTATION_GUIDE.md		IMPLEMENTATION_GUIDE.md
LICENSE		LICENSE
README.md		README.md
STEP_BY_STEP_GUIDE.md		STEP_BY_STEP_GUIDE.md
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation