Healthcare Assistant RAG Chatbot

Instant, accurate, and personalized healthcare advice by combining the NHS (National Health Services ) A-Z knowledge base with the power of LLM-driven Retrieval-Augmented Generation.

Comment

Healthcare Assistant RAG Model

Inspiration

I would often find myself on this website, going through symptoms or medical conditions I was experiencing, such as prolonged cold or cough, reading through medical advice and recommended medicines. The website contains more than 1,000 medical conditions, providing advice, self-care tips, treatment options, and more. This inspired me to build a RAG model around this knowledge base.

What It Does

The Healthcare Assistant RAG Model allows users to input healthcare-related queries and receive personalized, AI-enhanced responses. It retrieves relevant documents from the NHS A-Z knowledge base, processes the query alongside these documents, and generates an informed, contextually accurate reply.

How We Built It

We built the system in stages:

  1. Data Scraping: Scraped unstructured, textual data from the NHS A-Z website, which includes advice on medical conditions, possible treatments, and self-care routines.
  2. Database Setup: Stored the data in MongoDB Atlas with vector embeddings and created a vector search index.
  3. RAG Engine: Used LangChain to integrate relevant documents retrieved from MongoDB with an LLM to generate personalized responses, complete with actual sources of information.

Challenges We Ran Into

  • Scraping data from HTML pages with no defined structure for the text content.
  • Tying the RAG engine with a REST API.

Accomplishments That We're Proud Of

  • Building a full-stack, LLM-powered application.
  • Successfully working and collaborating as a team.
  • Leading a team to deliver the project.

What We Learned

  • Developing a complete RAG pipeline from data ingestion to response generation.
  • Overcoming challenges in web scraping and unstructured data handling.
  • Efficient use of embeddings and vector similarity search using MongoDB Atlas Vector Search.
  • Effective teamwork and leadership in a hackathon setting.

What's Next for Healthcare Assistant RAG Model

  • Enhance the web scraper to update data periodically from the NHS website.
  • Add multi-language support for a wider audience.
  • Incorporate additional healthcare datasets to improve the breadth of information.
  • Optimize the embedding generation and similarity search processes for faster responses.
  • Expand the functionality to include voice-based interactions and mobile app support.

Built With

Share this project:

Updates