My research focuses on language model evaluation, scientific exploration, and decision-making in agent systems.
Explore my researchWhat must an AI understand
to make its next move?
Extract the information and keep a link to where it came from.
Chart extraction · Biomedical retrievalConceptual illustrations of the research questions.
My early work asked how to recover information from scientific charts and connect biomedical answers to their sources. It led to a question that still guides me: what information does a system need to make a sound judgment?
Finding structure in informationExtract the information and keep a link to where it came from.
Chart extraction · Biomedical retrievalEvidence has to be checked and interpreted. Through multi-agent meta-analysis and studies of clinician trust, I began examining how agents and people combine, question, and use what they know.
From evidence to shared conclusionsAgents cross-check sources. People need evidence they can inspect and interpret.
AutoMETA · Clinician trustScientific exploration also involves acting: editing a composition or a sequence and observing what changes. My work asks whether those edits preserve a path to the goal—and whether our evaluation can see where that path is lost.
Explore patterns, edits, and possibilitiesBoth edits are allowed. Only one leaves a way to complete the task.
Evaluating LLM Agents Beyond Local Edit ValidityI now want to connect these questions through representation learning and world models: learning relationships that transfer, anticipating the effects of actions, and knowing when more information is needed.
Research directionsCan a learned model anticipate which edit would preserve a route to the goal?
Research direction · Representation learning & planningFrom changing a letter to editing a composition or protein sequence: I study whether a valid local change keeps a successful outcome within reach.
Agents & planningSmall validation sets can assign the same score to different checkpoints. I study what those ties hide about generalization and model selection.
Learning & evaluationStudy-centered agents extract, critique, and revise evidence before a statistical module synthesizes the results.
Multi-agent systemsAn exploratory study follows how people inspect, revisit, and interpret evidence while evaluating an AI-generated answer.
Human–AI interactionI work at the Human Artificial Intelligence Research Lab ↗ at Yonsei University, with Professor Keeheon Lee. My research brings together machine learning, agent systems, and the people who use them.
More about me