---
title: Medical Report Analyzer
emoji: 🏥
colorFrom: blue
colorTo: green
sdk: docker
app_file: app.py
pinned: true
---

# 🏥 AI Medical Report Analyzer

An intelligent multi-agent AI system that analyzes medical reports and explains them in simple Hindi or English — making healthcare accessible to every Indian.

---

## 🎯 Problem Statement

Every year, millions of Indians receive medical reports they don't understand. Blood reports, X-rays, prescriptions — full of technical jargon that patients can't decode. Most can't afford to visit a doctor just to explain what their report means.

This project gives every Indian access to an AI-powered medical assistant that reads their report, explains it simply, and tells them exactly what to do next.

---

## ✨ Features

- 📄 **Multi-format support** — PDF, JPG, PNG (photos, scans, digital reports)
- 🔬 **Abnormality detection** — flags every value outside normal range with severity
- 🧠 **RAG-powered diagnosis** — searches 2457 diseases from medical knowledge base
- 💬 **Simple explanations** — no medical jargon, patient-friendly language
- 👨‍⚕️ **India-specific advice** — AIIMS, government hospitals, specialist recommendations
- 🚦 **Risk levels** — LOW / MEDIUM / HIGH / CRITICAL with urgency timing
- 🇮🇳 **Hindi support** — full output in Hindi or English
- 📝 **Downloadable report** — color-coded Word document (.docx)
- ⚠️ **Responsible AI** — medical disclaimer on every response

---

## 🏗️ Architecture

```
User Input (PDF / Image)
         ↓
[Agent 1] Document Agent     — pypdf + Gemini Vision
         ↓
[Agent 2] Abnormality Agent  — Gemini 2.5 Flash
         ↓
[Agent 3] RAG Agent          — FAISS + sentence-transformers
         ↓
[Agent 4] Explanation Agent  — Gemini + deep-translator
         ↓
[Agent 5] Specialist Agent   — Gemini (India-specific)
         ↓
[Agent 6] Report Agent       — python-docx
         ↓
Gradio UI + Downloadable Report
```

All agents orchestrated by **LangGraph** with shared TypedDict state.

---

## 🛠️ Tech Stack

| Component | Technology |
|---|---|
| LLM + Vision | Gemini 2.5 Flash (free tier) |
| Agent Orchestration | LangGraph |
| Vector Database | FAISS (local) |
| Embeddings | sentence-transformers/all-MiniLM-L6-v2 |
| Medical Dataset | HuggingFace — kamruzzaman-asif/Diseases_Dataset |
| PDF Extraction | pypdf |
| Translation | deep-translator |
| Report Generation | python-docx |
| UI | Gradio |
| Deployment | HuggingFace Spaces |

**Cost: ₹0 — 100% free stack**

---

## 📊 Knowledge Base

| Source | Entries | Strength |
|---|---|---|
| IndianServers split | 796 | India-specific diseases |
| QuyenAnh split | 399 | Has treatment data |
| symptom2disease split | 1040 | Natural language symptoms |
| itachi9604 split | 222 | Additional coverage |
| **Total** | **2457** | **2535 FAISS chunks** |

---

## 🚀 Setup & Installation

### Prerequisites
- Python 3.11
- Gemini API key (free at [aistudio.google.com](https://aistudio.google.com))

### Installation

```bash
# Clone repository
git clone https://github.com/your-username/Medical-Report-Analyzer
cd Medical-Report-Analyzer

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

# Install dependencies
pip install -r requirements.txt
```

### Configuration

Create `.env` file:
```
GOOGLE_API_KEY=your_gemini_key_here
```

### Build Knowledge Base (run once)

```bash
python prepare_data.py    # Downloads and prepares medical dataset
python build_faiss.py     # Builds FAISS vector index
```

### Run

```bash
python app.py
```

Open `http://127.0.0.1:7861` in your browser.

---

## 📁 Project Structure

```
Medical-Report-Analyzer/
│
├── agents/
│   ├── document_agent.py      # PDF + image extraction
│   ├── abnormality_agent.py   # Detects abnormal values
│   ├── rag_agent.py           # FAISS semantic search
│   ├── explanation_agent.py   # Simple language explanation
│   ├── specialist_agent.py    # India-specific doctor advice
│   └── report_agent.py        # Word document generation
│
├── data/
│   └── medical_knowledge.txt  # Built from HuggingFace dataset
│
├── faiss_index/               # Vector index (built locally)
├── generated_reports/         # Patient reports saved here
├── graph.py                   # LangGraph pipeline
├── build_faiss.py             # One-time FAISS builder
├── prepare_data.py            # HuggingFace data loader
├── app.py                     # Gradio UI
├── config.py                  # Central configuration
└── requirements.txt
```

---

## 🔒 Safety & Disclaimer

This system is designed with responsible AI principles:

- Every response includes a medical disclaimer
- System never claims to diagnose — only guides
- Stronger disclaimers for imaging (X-ray, MRI)
- Always recommends consulting a qualified doctor
- Emergency helplines included in every report

---

## 🗺️ Roadmap

**V1 (Current)**
- Blood reports, prescriptions, discharge summaries
- Hindi + English
- 2457 disease knowledge base

**V2 (Planned)**
- Parallel RAG + web search agents
- Voice input support
- More Indian languages
- WhatsApp integration

---

## 👨‍💻 Author

Built by Bhavya as a portfolio project demonstrating:
- Multi-agent AI systems with LangGraph
- RAG pipelines with FAISS
- Multimodal AI with Gemini Vision
- Production-grade Python architecture