people

I am a second-year Ph.D. student in Computer Science at the University of Maryland (UMD), advised by Prof. Soheil Feizi. Broadly speaking, my research interests span problems related to the reliability of the evaluation and deployment of LLMs and LLM agents. More specifically, I have worked on reliable LLM evaluation (EMNLP’25), reliable agentic tool call in LLM agents (EMNLP’25), aligning LLM tool call behavior with human in multi-turn settings (ACL’26), and red-teaming AI text detection (NeurIPS’25). I’ve also worked on LLM reasoning, including interpretable reasoning (ACL’26) and explorative reasoning (arXiv’26).

Prior to joining UMD, I earned my bachelor’s degree in Computer Science and Electrical Engineering from the Hong Kong University of Science and Technology (HKUST). During my undergraduate studies, I had the opportunity to work with Prof. Minhao Cheng on backdoor attacks and machine learning watermarks. I also spent a semester on exchange at ETH Zurich, where I was fortunate to have worked with Prof. Florian Tramèr on adversarial examples and diffusion models.

I am a second-year Ph.D. student in Computer Science at the University of Maryland (UMD), advised by Prof. Soheil Feizi. Broadly speaking, my research interests span problems related to the reliability of the evaluation and deployment of LLMs and LLM agents. More specifically, I have worked on reliable LLM evaluation (EMNLP’25), reliable agentic tool call in LLM agents (EMNLP’25), aligning LLM tool call behavior with human in multi-turn settings (ACL’26), and red-teaming AI text detection (NeurIPS’25). I’ve also worked on LLM reasoning, including interpretable reasoning (ACL’26) and explorative reasoning (arXiv’26).

Prior to joining UMD, I earned my bachelor’s degree in Computer Science and Electrical Engineering from the Hong Kong University of Science and Technology (HKUST). During my undergraduate studies, I had the opportunity to work with Prof. Minhao Cheng on backdoor attacks and machine learning watermarks. I also spent a semester on exchange at ETH Zurich, where I was fortunate to have worked with Prof. Florian Tramèr on adversarial examples and diffusion models.