Research scope
Compared biomedical language and vision-language models, adapted MEDITRON with low-rank fine-tuning, built biomedical retrieval pipelines, and studied specialized agents for diagnosis. The work evaluated both benchmark accuracy and failure modes in complex clinical cases.
A connected research program
This internship and the master’s thesis cover the same medical AI research program. The detailed research page documents the training configuration, retrieval design, agent ablations, evaluation metrics, and limitations.