I am an Assistant Professor of Surgery at the University of Minnesota, where I develop interpretable computational methods to identify clinically meaningful states from sparse, real-world longitudinal data. My research lies at the intersection of translational data science, clinical inference, and multi-omics integration, with a focus on how pain, recovery, and disease burden evolve across time in complex human disease.

My work is grounded in the idea that many clinically important outcomes like  persistent pain, incomplete recovery, treatment response, and disease progression  are dynamic processes rather than static traits. These processes are shaped not only by biology, but also by treatment, monitoring, and care context, even as the data used to study them remain irregular, incomplete, and deeply conditioned by how healthcare is delivered. I address this challenge by building computational frameworks that are predictive, interpretable, and biologically grounded, while remaining attentive to the constraints of real-world data generation.

Across clinical and translational settings, I use longitudinal clinical data, patient-reported outcomes, and molecular measurements to define latent disease and recovery states. A particular focus of my research is observability: what healthcare data truly allow us to infer, what they obscure, and how workflow, digital infrastructure, treatment intensity, and follow-up shape the patterns that models learn. This perspective informs my work on validation, transportability, and context-aware clinical AI.

Through projects spanning pain and recovery phenotyping, multimodal integration, and translational decision support, I aim to move beyond static endpoints and black-box prediction toward computational approaches that more faithfully capture the temporal, mechanistic, and context-dependent nature of human disease.

Research focus

My research program focuses on three closely linked areas:
1.Dynamic phenotyping: To identify clinically meaningful states of pain, recovery, and disease burden over time.
2.Observability and translation validity: To understand when models and phenotypes remain interpretable across datasets and health systems.
3.Interpretable multimodal integration: To connect longitudinal clinical signals with biologic heterogeneity in ways that support translational insight rather than prediction alone.