AI Research Scientist at Lunit working on reliable foundation models for biomedical AI.
My research studies how learned systems generalize, adapt, and fail when their data or evidence changes. I work across representation learning, LLM post-training, retrieval, and evaluation.
Clinical RAG can faithfully cite real evidence while attributing it to the wrong medical entity. We characterized this failure across 13 models and a deployed system, then proposed entity-attribution verification.
A transformer for cross-dataset Cell Painting representations that generalizes to unseen datasets without fine-tuning.
Paper · Code · Project page
Reliable adaptation · Distribution shift · Medical foundation models · Evidence-grounded reasoning · Model evaluation
Website · Google Scholar · LinkedIn · Email
In the fields of observation, chance favors only the prepared mind.