Self-Evolving Agents via Likelihood-Guided Tool-Space Optimization

LOTS uses observed tool outputs to learn reusable, task-specific tool selections that improve agent performance while reducing tool context.

LatentGym: A Testbed For Cross-Task Experiential Learning With Controllable Latent Structure

LatentGym uses environments with known latent structure to measure how LLM agents gather information and adapt across related tasks.

Research Scientist

Decision, Risk, and Operations division at Columbia Business School, working with Prof. Hongseok Namkoong on continual, meta, and reinforcement learning.