🧠 AI Healthcare Multi-Agent System

A modular AI-powered healthcare assistant built using LLMs + LangChain, designed to provide structured, personalized insights based on disease detection and patient data.

---

🚀 Overview

This project is not a general chatbot. It is a multi-agent system that:

• Takes structured medical JSON input
• Uses specialized AI agents for different tasks
• Produces schema-constrained outputs
• Integrates ML disease detection + LLM reasoning

---

🧩 System Architecture

User Input (Medical JSON / Image)
↓
Disease Detection Models
↓
Structured JSON Output
↓
LangChain Agent Router
├── Lifestyle Suggestor Agent
├── Hospital Finder Agent (with external API)
├── Cost Estimator Agent
└── Diet Planner Agent
↓
Response Aggregator
↓
Final Structured Output

---

🤖 AI Agents

1. 🧘 Lifestyle Suggestor Agent

• Generates personalized lifestyle recommendations
• Focus: habits, sleep, exercise, stress

2. 🏥 Hospital Finder Agent

• Finds relevant hospitals based on disease & location
• Can integrate with APIs (Google Maps / SerpAPI)

3. 💰 Cost Estimator Agent

• Provides approximate treatment cost (INR)
• Includes cost breakdown and influencing factors

4. 🥗 Diet Planner Agent

• Generates disease-specific diet plans
• Supports Indian food context and dietary preferences

---

🧪 Disease Detection Models

The system integrates pretrained deep learning models for automated disease prediction:

🫁 Pneumonia Detection

• Model: AurevinP/pneumonia-classifier-effnetb0
• Architecture: EfficientNet-B0
• Link: https://huggingface.co/AurevinP/pneumonia-classifier-effnetb0

👁️ Diabetic Retinopathy (DR)

• Model: rafalosa/diabetic-retinopathy-224-procnorm-vit
• Architecture: Vision Transformer (ViT)
• Link: https://huggingface.co/rafalosa/diabetic-retinopathy-224-procnorm-vit/tree/main

🧠 Alzheimer Detection

• Model: subx24/ml-alzheimer-models
• Architecture: VGG19
• Link: https://huggingface.co/subx24/ml-alzheimer-models/tree/main

---

⚙️ Tech Stack

• LLM: Gemini API
• Framework: LangChain
• Backend: Node.js / Python (extensible)
• ML Models: HuggingFace Transformers
• Data Format: Structured JSON

---

📦 Input Format

All agents consume a standardized JSON input:

{
  "patient": {
    "age": 45,
    "gender": "male"
  },
  "disease": {
    "name": "Type 2 Diabetes",
    "severity": "moderate"
  },
  "location": {
    "city": "Chandigarh"
  },
  "preferences": {
    "diet_type": "vegetarian",
    "budget": "medium"
  }
}

---

▶️ Run Locally

Prerequisites

• Node.js
• Gemini API Key

Steps

1. Install dependencies:

   npm install

2. Set your API key in .env.local:

   GEMINI_API_KEY=your_api_key_here
   

3. Run the app:

   npm run dev

---

🎯 Key Features

• ✅ Multi-agent architecture (not a chatbot)
• ✅ Structured medical reasoning
• ✅ Modular and scalable design
• ✅ Integration of ML + LLM
• ✅ Real-world healthcare use-case

---

🧠 How This Differs from ChatGPT

Feature	ChatGPT	This System
Input Type	Unstructured text	Structured JSON
Output	Free-form	Schema-constrained
Architecture	Single model	Multi-agent system
External Tools	Limited	Integrated APIs
Domain Focus	General	Healthcare-specific

🔮 Future Improvements

Real-time hospital API integration

Patient history tracking

Risk prediction models

Doctor recommendation ranking system

⚠️ Disclaimer This system is for educational and assistive purposes only and does not replace professional medical advice.

📌 Author Developed as part of an AI healthcare system project using modern LLM orchestration techniques.