I am a final-year Ph.D. candidate in Electrical & Computer Engineering at Duke University, advised by Prof. Maciej A. Mazurowski. I develop foundation models and agentic AI systems that reason across images and language, use specialized tools, and adapt to complex real-world tasks.

My work spans LLM-agent optimization at Meta, in-context 3D medical image segmentation at Siemens Healthineers, and large-scale multimodal learning at Duke. I previously earned my M.S. in Computer Science at Duke, where I worked with Prof. Guillermo Sapiro, and B.S. degrees in Mathematics–Computer Science and Cognitive Science at UC San Diego, where I worked with Prof. Zhuowen Tu.

I am seeking research scientist, applied scientist, and research engineering positions starting in Summer 2027.

Download résumé

Selected Work

Meta-Harness — LLM agent optimization

Improved Meta-Harness’s algorithm-evolution pipeline by evolving agent trajectories and dynamically adapting training data. Increased performance from 49% to 62% on math-solving tasks and from 68% to 71.3% on Terminal-Bench 2.

In-context 3D segmentation of unseen anatomies

Developed a 3D nnU-Net framework with multi-stage cross-attention that segments new anatomical structures from annotated support examples without retraining. Curated 14,934 CT volumes spanning 92 structures and achieved 52.1% Dice on held-out unseen anatomies, compared with 36.7% for the strongest prior in-context learning baseline. A systematic support-selection study revealed a 21-point Dice performance range. Co-inventor on the resulting patent-pending technology assigned to Siemens Medical Solutions USA, Inc.

MRI-CORE — MRI foundation model

Pretrained a foundation model on 7M MRI slices, improving few-shot downstream segmentation by 5 Dice points over SAM, and extended this direction toward multimodal 3D MRI and radiology-report alignment.

MMedAgent — Multimodal medical agent

Developed a multimodal agent that coordinates specialized medical tools across imaging and language tasks; published in Findings of EMNLP 2024.

Experience

Software Engineering Intern, Meta · May 2026 – Present
Worked on algorithm evolution for LLM agents, making trajectories evolvable and adapting training data to strengthen optimization signals.

AI Research Intern, Siemens Healthineers · June 2025 – August 2025
Developed a 3D in-context segmentation framework for unseen anatomies, built and evaluated on 14,934 CT volumes across 92 structures; the work resulted in patent-pending technology assigned to Siemens Medical Solutions USA, Inc.

Recent News

• [Sep. 2026] Our work on learning CT representations across anatomy and disease was accepted to npj Digital Medicine.
• [Sep. 2026] Our work on muscle segmentation across anatomical regions in MRI was accepted to Computers in Biology and Medicine.
• [Jan. 2026] Our work on Fréchet Radiomic Distance was published in Medical Image Analysis.
• [Jan. 2026] Our work applying SAM 2 to 2D and 3D medical images was published in IEEE Transactions on Biomedical Engineering.
• [Dec. 2025] Our work on breast MRI registration was published in IEEE Journal of Biomedical and Health Informatics.
• [Oct. 2025] Our work on AI-based breast density quantification was published in npj Breast Cancer.
• [Oct. 2025] Our work on volumetric annotation with SAM 2 was published in IEEE Transactions on Medical Imaging.
• [July 2025] We released MRI-CORE, a foundation model for magnetic resonance imaging.