About

I am a final-year PhD student in Machine Learning in the H. Milton Stewart School of Industrial & Systems Engineering (ISYE) at Georgia Institute of Technology (GT), being advised by Dr. Gian-Gabriel Garcia and Dr. Kamran Paynabar.

Job Market Paper: Constraint-Aware Self-Improving Large Language Model for Clinical Role Model Generation. Preprint available on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5642250

• We model case-based personalized medicine as a computationally tractable chance-constrained optimization program that incorporates uncertainties in risk estimates from historical treatment trajectories
• We use LLMs to generate candidate treatment trajectories, with the chance-constrained program serving as an external verifier to label reliability of synthetic trajectories
• We develop an active learning algorithm that strategically fine-tunes the LLM based on the verification feedback
• We establish a complete theoretical framework with guarantees for single-step LLM alignment, learning algorithm regret bounds, and patient safety
• On two large clinical datasets, our framework significantly outperforms both real treatment trajectories and active learning benchmarks, improving LLM reliability by over 130% while reducing patient Effort-to-Change by 23%
• We present a provably effective paradigm for using LLMs to overcome data scarcity in data-driven optimization, which is applicable beyond clinical settings to any optimization model where data approximates decision variables