Experience

01

Siemens Healthineers

Jun – Aug 2026 · Princeton, NJ

AI Medical Imaging Intern

• Built an automated 3D U-Net/ResNet oriented-bounding-box detection pipeline for surgical imaging — 99.6% vertebral-center landmark accuracy.
• 2.3 mm mean localization error and 3.5° median rotation error on held-out volumetric scans.
• Owned the full ML pipeline end-to-end: preprocessing, augmentation, training, tuning, evaluation.

02

Duke University

May 2026 – Present · Advisors: Dr. Pan Xu, Weixin Wang

Undergraduate Researcher · Deep RL for Portfolio Optimization

• Formalized allocation as a regime-switching POMDP with a causal HMM filter over VIX and realized volatility.
• Multivariate Student-t return models with Bayesian shrinkage to capture fat tails and crisis correlations.
• Extended PPO with belief-state observations and regime-aware Differential Sharpe reward shaping.