Medical Chatbot
A medical question answering assistant. BERT handles retrieval and understanding, Llama 2 rewrites the result into an answer.
- 2025
- Personal project
- Natural language processing
Overview
The chatbot answers medical questions using two models with different jobs. A BERT model trained on medical data does the understanding and retrieval, finding the relevant material for a question.
Llama 2 then reformulates that material into a readable answer. Splitting the work this way keeps the answer anchored to retrieved content rather than generated from scratch.
I built this to learn how the two model types behave in practice, and where each one is the wrong tool for the job.
What I built
- Prepared the medical dataset and fine tuned BERT on it
- Built the retrieval step and the scoring around it
- Wired Llama 2 in to reformulate retrieved passages
- Evaluated answers against a held out set of questions
Smart Library System