Projects

Exploring methods to improve reward quality and density for more reliable optimization in verifiable reward settings.

Improving medical LLM benchmark performance through systematic prompt optimization and reliability-oriented evaluation.

Investigated why deep GNNs lose performance at depth and found training dynamics drive oversmoothing more than architecture limits.

Built a full-stack AI fashion assistant that recommends outfits from existing wardrobes and visualizes looks with virtual try-on.

Enabled natural language access to manufacturing databases using an NL2SQL agent with validation, execution, and conversational memory.

Created a Python-based 3D reconstruction pipeline that recovers facial geometry from one 2D image with BFM and FLAME models.

Developed a web app that helps users find medically accurate information and generate structured reports with AI support.

Built a ChatGPT-style app with real-time conversation, authentication, and state management using React and Node.js.

Built a platformer game in Python using the Pygame library.