README.md

metadata

title: Epilepsy Management AI Agent
emoji: 🧠
colorFrom: red
colorTo: purple
sdk: docker
app_port: 7860
tags:
  - streamlit
  - medical-ai
  - epilepsy
  - chromadb
  - semantic-search
  - healthcare
pinned: false
license: apache-2.0
short_description: AI-powered epilepsy device recommendations

🧠 Epilepsy Management AI Agent

AI-powered medical device recommendation system for epilepsy management using semantic search, safety filtering, and evidence-based ranking.

🎯 What This App Does

This Streamlit application helps clinicians and patients explore epilepsy management devices based on patient-specific descriptions. It uses:

• Semantic Search: ChromaDB + Sentence Transformers for intelligent device matching
• Safety Rules: Configurable filters for invasiveness and regulatory approvals (FDA/CE)
• Interactive UI: Multi-page interface with device catalog, admin tools, and recommendations

✨ Features

• 📝 Patient Input → Smart Recommendations: Describe patient condition and receive personalized device suggestions
• 🔍 Semantic Search: Uses all-MiniLM-L6-v2 embeddings for intelligent text matching
• 🛡️ Safety Filters: Toggle invasive device filtering and requirement for specific approvals
• 📚 Device Catalog: Browse and search the complete knowledge base
• 🛠️ Admin Tools: Upload CSV data and rebuild vector index
• 📊 Detailed Results: View device specifications, approvals, indications, and source links

🚀 How to Use

1. Enter Patient Description: Describe the patient's epilepsy characteristics (seizure type, frequency, age, lifestyle needs)
2. Configure Filters:
   • Toggle "Avoid invasive devices" on/off
   • Specify required approvals (e.g., "FDA, CE")
   • Adjust top K results (1-10)
3. Generate Recommendations: Click the button to get AI-filtered device suggestions
4. Explore Results: View device details, click source links, and browse the full catalog

📋 Example Query

"16-year-old with focal aware seizures 2–3 times/week, mostly nocturnal; wants non-invasive detection and caregiver alerts."

🏗️ Architecture

Patient Input → Semantic Search (ChromaDB) 
             → Safety Rules Filtering 
             → Ranked Recommendations 
             → Interactive Display

Technologies

• Streamlit: Web UI framework
• ChromaDB: Vector database for semantic search
• Sentence Transformers: Text embeddings (all-MiniLM-L6-v2)
• Pandas: Data manipulation
• Python 3.11: Runtime environment

📁 Project Structure

Epilepsy-Management-AI-Agent/
├── src/
│   └── app.py              # Main Streamlit application
├── pages/
│   ├── 1_📚_Device_Catalog.py    # Browse all devices
│   ├── 2_🛠️_Admin_ETL.py        # CSV upload & index rebuild
│   └── 3_ℹ️_About.py            # Project information
├── data/
│   └── devices.csv          # Device catalog database
├── utils/
│   ├── embeddings.py        # ChromaDB integration
│   └── rules.py            # Safety rule filtering
├── requirements.txt         # Python dependencies
├── Dockerfile              # Container configuration
└── README.md              # This file

⚙️ Deployment

This app is containerized using Docker and deployed on Hugging Face Spaces.

Local Development

# Build and run
docker build -t epilepsy-ai .
docker run -p 7860:7860 epilepsy-ai

# Or with Streamlit directly
pip install -r requirements.txt
streamlit run src/app.py

📊 Data Model

The device catalog (data/devices.csv) includes:

• device_id: Unique identifier
• name: Device name
• category: Device type (wearable, implant, etc.)
• indication: Use case
• invasiveness: Invasive/Non-invasive
• approvals: Regulatory status (FDA, CE, etc.)
• summary: Description text
• source_url: Reference link
• last_seen: Timestamp

⚠️ Important Notice

This tool provides informational recommendations only and is not a substitute for professional medical advice. Always consult a licensed clinician for diagnosis and treatment decisions.

🔧 Configuration

• ChromaDB: Persists in .chroma/ directory (rebuilds automatically on HF Spaces)
• Model: sentence-transformers/all-MiniLM-L6-v2 (~80MB, downloads on first run)
• Port: 7860 (Hugging Face default)

📝 License

Apache 2.0

👥 Contributing

Feel free to submit issues or improvements to the device catalog or codebase.

---

Built with ❤️ for epilepsy management and healthcare innovation.