Research agenda

Trustworthy personalized
health agents

From multimodal data to verifiable action. My research focuses on AI systems that build an understanding of the individual, interpret changes in health over time, and connect their responses and proposed actions to evidence. I combine agent architectures, knowledge graphs, and causal reasoning to advance personalized health technology.

01 / PERSONALIZATION

Understanding
the individual

Personalization requires more than adapting the wording of a response. An agent needs relevant context about the person and a way to recognize what is missing.

I study how agents can acquire, structure, and update patient context using clinical knowledge, personal history, and observations over time. This includes targeted clarification, context-aware reasoning, and coordination between agents for personalized care.

Current work: Knowledge-guided clarification and Personal Care Utility for type 2 diabetes.
02 / MULTIMODAL UNDERSTANDING

Health data
& event mining

Signals and clinical records capture different aspects of health. Understanding them together means identifying meaningful events and the relationships between them.

I work with wearable and physiological signals alongside clinical text to build interpretable representations of health. My research includes signal analysis, physiology-guided causal discovery, and event-graph mining to detect changes, examine event relationships, estimate next-event probabilities, and interpret their clinical meaning.

Current work: Multimodal event analysis and interpretable health modeling, with a causal-discovery study evaluated across two cohorts totaling 2,365 participants.
03 / VERIFIABLE ACTION

Evidence for
responses & actions

A health agent should be able to show what supports its response, recommendation, or proposed action, and identify when that support is incomplete.

I develop methods to connect LLM outputs to structured evidence and explicit requirements. My work checks the support behind generated responses; the broader agenda extends this to agent decisions and tool-mediated actions, with traceable evidence and constraint checks before execution.

Current foundation: Knowledge graphs and evidence-gated LLM evaluation. Research direction: Verification of proposed actions within personalized health-agent workflows.

Systems & applications

Connecting research to health technology

These directions inform my work on personalized health agents, analysis of wearable data, and clinical reasoning systems. The aim is to make patient context, detected events, and supporting evidence usable within a coordinated AI workflow.

I contribute through data preparation, architecture design, implementation, and evaluation. My experience also includes clinical-trial matching, care-companion robotics, and automotive sensing.

Get in touch

Building trustworthy AI for health? Let’s talk.

I’m interested in research roles in health AI, LLM agents, and wearable and clinical data.