Experience

worked with high-frequency order-flow data to search for predictive signals in a constrained feature set. i trained LightGBM models to predict rapidly filled orders and estimate their resulting hedged price impact on the underlying asset. alongside the research work, i applied options theory, forecasting, and risk management in mock trading and placed 5th out of 60 interns.

conducting robotics research under sergey levine and william chen. i built an inference pipeline that periodically queries Gemini over visual observations, producing language-conditioned actions for closed-loop robotic manipulation with steerable policies. i’m currently exploring human-feedback distillation methods that reformulate a policy’s state and action spaces to better incorporate corrective feedback during training.

supported the development of ai and llm systems through structured model evaluation, data-quality analysis, and feedback. i collaborated with research teams to identify failure modes and improve model behavior across a range of tasks and domains.

built CourseMate, a full-stack academic-management platform using Next.js, React, TypeScript, and Flask. i designed PostgreSQL schemas, developed REST APIs with JWT authentication and role-based access control, added file storage with AWS S3, and helped deploy the platform using EC2, RDS, and CI/CD pipelines.