title	MedBot
emoji	🤖
colorFrom	blue
colorTo	green
sdk	gradio
sdk_version	6.15.2
python_version	3.11
app_file	app.py
pinned	false

medbot

A retrieval-augmented generation (RAG) demo project designed for a Hugging Face Spaces-friendly Gradio deployment.

Hugging Face Spaces Ready

This repository is structured for a Space that uses a Gradio app. To make the Space reproducible, the project includes:

• requirements.txt with a pinned Gradio version.
• README.md with installation, usage, and deployment notes.
• ui/gradio_dashboard.py as the intended Gradio app entrypoint.

Requirements

Recommended environment:

• Python 3.10 or 3.11
• gradio==6.15.2
• numpy<2
• requests
• PyPDF2==3.0.1
• sentence-transformers==5.5.1
• faiss-cpu==1.8.0
• torch>=2.0.0
• python-dotenv for loading local .env files when present

Install dependencies:

pip install -r requirements.txt

If faiss is not available in your local environment, the pipeline will still work with a fallback search method, but installing faiss-cpu gives better retrieval performance.

Local Run

If you add your Gradio interface to ui/gradio_dashboard.py, run locally with:

python ui/gradio_dashboard.py

For a basic Gradio app, this file should create a gr.Interface or gr.Blocks instance and call launch().

Standalone Retrieval Script

A simple root-level script is available for running retrieval directly from a pickle store:

python main.py --query "what is diabetes" --top-k 3

By default, the script loads rag_store_cpu.pkl from the repository root. To specify a different store file:

python main.py --path path/to/rag_store_cpu.pkl --query "what is diabetes"

Deployment Notes for Hugging Face Spaces

To deploy on Hugging Face Spaces:

1. Push this repo to a Hugging Face Space.
2. Ensure requirements.txt is present.
3. Confirm app.py or your Gradio entrypoint is referenced in the Space settings if needed.
4. Use a compatible runtime: CPU is fine for light demos, GPU may be required for larger models.

Recommended Space settings

• Runtime: cpu for simple demos
• Python version: 3.10 or 3.11
• Install command: pip install -r requirements.txt

What to Add

This repo currently has a working app shell at ui/gradio_dashboard.py.

The app is already configured to load these medical PDF links:

• https://bookchapter.org/kitaplar/Medical%20Diagnosis%20and%20Treatment%20Methods%20in%20Basic%20Medical%20Sciences.pdf
• https://applications.emro.who.int/dsaf/dsa236.pdf
• https://www.dhhs.nh.gov/sites/g/files/ehbemt476/files/documents/2021-11/disease-handbook-complete.pdf
• https://aiimsrishikesh.edu.in/documents/standard-treatment-guidelines.pdf
• https://www.fao.org/4/y1238e/y1238e06.pdf

To use the demo, configure the Gemini API key on the backend, then run the app locally:

export GEMINI_API_KEY="your_gemini_api_key"
python ui/gradio_dashboard.py

Supported backend key names: GEMINI_API_KEY, GOOGLE_API_KEY, and GOOGLE_GENAI_API_KEY.

Then open the Gradio link in your browser, ask a question, and the backend will retrieve PDF passages before sending them to Gemini.

License and Attribution

Add a license file and any model attribution required by your dataset or API provider.

Notes

If you need a direct model or API integration, add the dependency to requirements.txt and document it here.