Seongheon Park

Hello! I’m a third-year Ph.D. student in the Department of Computer Sciences at the University of Wisconsin-Madison, where I am fortunate to be advised by Prof. Sharon Li. Previously, I received my M.S. and B.S. degrees in Electrical and Electronic Engineering from Yonsei University, where I was advised by Prof. Kwanghoon Sohn and Prof. Kibok Lee.

I am actively seeking Summer 2027 research internship opportunities, particularly in foundation models, AI agents, and robotics. Please get in touch!

My research focuses on reliable foundation models (LLMs, LVLMs, and VLAs) and self-improving AI agents, with applications in language, multimodal reasoning, and robotics. I study why and how models and agents fail, and how signals from their behavior and internal representations can help them evaluate decisions, recover from mistakes, and learn from experience.

My work connects three directions:

• Understanding failures: failure detection, uncertainty estimation, and interpretability to identify when and why models go wrong.
• Improving decisions at test time: verification, progress evaluation, and test-time scaling for language and robotic agents.
• Post-training and adaptation: understanding and improving how agents use feedback to refine their behavior.

My broader goal is to enable agents to complete long-horizon tasks reliably in unfamiliar environments, recognize when to replan or seek human assistance, and improve through interaction. I am particularly interested in how learned policies, memory, and runtime monitoring can work together within agent systems to support these capabilities.