Making Medical AI Smarter, Adapting Language Models for Real-World Healthcare

This international project brings together researchers from Canada and Japan to improve the accuracy and reliability of artificial intelligence (AI) in healthcare. The goal is to enhance how large language models (LLMs)—the technology behind tools like ChatGPT—respond to medical questions by grounding them in verified facts.

At the National Institute of Informatics (NII) in Tokyo, the team is building a Japanese medical language model using real clinical data and a technique called a Mixture-of-Experts. To ensure accuracy, the model will be linked to knowledge graphs, structured databases that help the AI provide more factual and explainable responses. This is especially important in healthcare, where misinformation can have serious consequences. Furthermore, this project will also address how to adapt medical AI systems across different languages and cultures.

The technology developed through this collaboration will directly support and advance the MARVIN chatbot project at Polytechnique Montréal, which helps people living with HIV manage their health. By applying these new tools, MARVIN will become more accurate, culturally aware, and effective for diverse users around the world.

Faculty Supervisor:

Sofiane Achiche;Bertrand Lebouché

Student:

Partner:

National Institute of Informatics

Discipline:

Engineering

Sector:

Education

University:

Polytechnique Montréal

Program: