Linhai Ma

I am a postdoc at Yale. My research focuses on making large language models reliable in tasks that require structured outputs, grounding in source evidence, and accurate evaluation of agent actions. I address these challenges through token-level preference optimization for structured generation, hypothesis-based retrieval, and reference-guided evaluation of agent tool calls, using clinical and financial text as primary testbeds. My work on financial agents is conducted in collaboration with The Fin AI Community.

Before joining Yale, I was a Machine Learning Engineer at Cadence Design Systems, where I worked on retrieval and reliability in an enterprise LLM-based question-answering system for chip design.

I received my Ph.D. in Computer Science from the University of Miami under the supervision of Dr. Liang Liang, where I worked on adversarial robustness of deep neural networks, with applications to medical image analysis.

I received my M.S. from the Institute of Software, Chinese Academy of Sciences, under the supervision of Dr. Peng Wu, where I worked on testing concurrent software systems.

My CV is available here.