Medical RAG-using-Meditron-7B-LLM

Oncology RAG Q&A App using Meditron 7B LLM, Qdrant Vector Database, and PubMedBERT Embedding Model.

Medical RAG QA App

Description

This FastAPI application integrates transformer-based language models (LLMs) and Retrieval-Augmented Generation (RAG) techniques to deliver precise medical question-answering. It utilizes SentenceTransformer embeddings and the Qdrant vector database to optimize user experience.

Data Ingestion of PDF Documents

Steps to Run

1. Setup and Data Ingestion

Before running the ingestion script, ensure Docker is installed and running on your machine.

1. Download Qdrant Image:

   docker pull qdrant/qdrant

2. Run Qdrant Container:

   docker run -p 6333:6333 qdrant/qdrant

3. Ingest Data: After setting up Qdrant, run ingest.py to create a vector database in Qdrant. Vectors are ingested into Qdrant via its API.

   python ingest.py

2. Semantic Similarity Search

Use retriever.py to perform semantic similarity search on a Qdrant vector database using sentence embeddings.

1. Run the Retriever:

   python retriever.py

3. RAG Application

Ensure you are logged into Hugging Face to access the necessary models.

1. Login to Hugging Face:

   huggingface-cli login

   Log in using the write token generated in your Hugging Face account.

   or directly Download the model from Hugging Face - meditron-7b-Q4_K_M-GGUF

2. Run the RAG Application: From the src folder, start the FastAPI application using Uvicorn:

   uvicorn rag:app

File Overview

• ingest.py: Ingests PDF documents and creates vector embeddings using SentenceTransformers. The vectors are stored in the Qdrant vector database.
• retriever.py: Uses LangChain to perform semantic similarity searches on the Qdrant vector database using sentence embeddings.
• rag.py: Implements the RAG (Retrieval-Augmented Generation) application using FastAPI. Connects to the Qdrant server using the specified URL.

Output

Additional Notes

• Ensure Docker is installed and running before starting the Qdrant container.
• Ensure you have the necessary permissions and tokens for Hugging Face.

Conclusion

This application leverages cutting-edge AI technologies to provide precise and efficient medical question-answering. By integrating LLMs, RAG, SentenceTransformer embeddings, and the Qdrant vector database, it ensures optimized performance and user experience.