Medical Rag Chatbot

I built Medical_RAG_Chatbot as a medical question-answering project on my GitHub account, github profile. It combines retrieval-augmented generation with local retrieval data and a Gradio interface.

My Medical_RAG_Chatbot repository is a Jupyter Notebook project with 1 star; it started 2026-03 and was last updated 2026-06.

The project stack is MedQuAD dataset, FAISS vector search, Sentence Transformers embeddings, Gemini models via LangChain, Gradio web UI.

The chatbot provides Medical Q&A grounded in MedQuAD context.

The app performs FAISS retrieval over local index files.

• The app supports Automatic model fallback on Gemini rate-limit errors.
• I can use Manual model switch from UI dropdown when choosing a model.
• The app uses Response caching to reduce repeated API calls.

The project files are app.py, requirements.txt, .env, medquad_2000.csv, faiss_index/index.faiss, faiss_index/index.pkl, deployment_package copies, and Medical_RAG_Chatbot (1).ipynb.

I need Python 3.10+ and a valid Gemini API key to run the project.

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

GEMINI_API_KEY=your_api_key_here
# Optional
# GOOGLE_API_KEY=your_api_key_here
# PORT=7865
# FAISS_INDEX_DIR=./faiss_index
# DATASET_PATH=./medquad_2000.csv

python app.py

Optional environment settings are GOOGLE_API_KEY, PORT=7865, FAISS_INDEX_DIR=./faiss_index, DATASET_PATH=./medquad_2000.csv.

If files are missing, the README says: If missing, copy from deployment_package/.

Copy-Item -Path ".\deployment_package\faiss_index" -Destination ".\faiss_index" -Recurse -Force
Copy-Item -Path ".\deployment_package\medquad_2000.csv" -Destination ".\medquad_2000.csv" -Force

The local app address is http://localhost:7865, or the PORT I set.

Setting BENCHMARK_RETRIEVAL to 1 makes startup log average, median, min, and max FAISS retrieval times at startup; no measured timing values are provided.

$env:BENCHMARK_RETRIEVAL="1"
python app.py

import os
os.environ["BENCHMARK_RETRIEVAL"] = "1"

!BENCHMARK_RETRIEVAL=1 python app.py

The deployment configuration is documented with a redacted server_name, a server port read from PORT with 7865 as its default, and sharing disabled.

pip install -r requirements.txt
python app.py

For Hugging Face Spaces, the setup is SDK: Gradio; App file: app.py; add GEMINI_API_KEY in Space secrets; upload faiss_index/ and medquad_2000.csv.

For credential security, .env is ignored via .gitignore; do not commit API keys.

• If the port is in use, set PORT=7866 and restart when the port is in use.
• For quota or rate-limit errors, use model switch dropdown or wait for quota reset.
• For missing data or index files, copy missing index/data files from deployment_package/.

The README also describes deployment on Render or Railway using pip install -r requirements.txt as the build command and python app.py as the start command. Deployment requires the Gemini key and dataset and FAISS index files. The exact Gemini models and fallback order, retrieval parameters, cache implementation, dataset size and preprocessing, measured timings, and deployment URL are not specified.