I am a second-year Ph.D. student in Computer Science at The University of Texas at Dallas, co-advised by Prof. Feng Chen and Prof. Bingzhe Li. My current focus is personal AI and physical AI, and the agentic memory that both rely on. Before this, I worked on computer-use agents (CUAs), including error recovery in long-horizon tasks and joint reinforcement learning of multiple CUAs. My work on personal agents is in collaboration with MatrAIx.
My research background spans LLM-based planning, structured reasoning for scientific writing, and dynamic text-attributed graphs. I was previously advised by Dr. Jessica Ouyang at UT Dallas. I have also worked with Sheng Li at the University of Virginia and Wei Xu at Renmin University of China.
I received my M.S. in Computer Science from Emory University, where I was advised by Prof. Fei Liu, and bachelor's degrees in Data Science and Finance from Renmin University of China.
I am looking for a Research Scientist Intern position for Summer 2027, and I am eligible for CPT. I also welcome research conversations and collaborations in these areas. Feel free to get in touch!
With the MatrAIx team, we released PersonalAgentEval, the first benchmark for personal agents across diverse users, personal preferences, and domains. We will grow it into an open leaderboard that anyone can contribute to, together with a large collaborative paper. Interested? You are welcome to join! (X · LinkedIn)
Three papers submitted to ACL ARR October 2026: PersonaCUABench (co-first author), SAFE-Mem (first author), and PersonaScienceBench.
I am looking for a Research Scientist Intern position for Summer 2027 (CPT eligible).
Two papers on computer-use agents are under review: CUA-Recovery (ICLR 2027) and our survey Toward Trustworthy Computer-Use Agents (ACM Computing Surveys).
Our paper, Enhancing Dynamic Text-Attributed Graph Modeling via Instruction Tuning, has been accepted to Findings of AACL-IJCNLP 2026!
I am serving as a reviewer for NeurIPS 2026 and IEEE BIBM 2026.
GRASP appeared in Findings of ACL 2026.
I started my Ph.D. in Computer Science at UT Dallas.
A benchmark of 44 personas across 11 domains that holds each task fixed and varies only who asks. Across nine CUA backbones, persona-phrased requests cut mean task completion from 0.689 to 0.441, and even fully completed requests satisfy users less often than the original requests.
An agentic memory that accounts for both the stored content and its information source. It has the lowest attack success among defenses that keep legitimate memory and the highest answer accuracy of all evaluated defenses.
A survey of trustworthiness in computer-use agents, organized by trust boundary and by how risks propagate after a failure.
Combining LLM planning and graph reasoning to capture relationships among papers and generate faithful related-work sections.
A review of planning benchmarks for LLMs across embodied environments, web navigation, scheduling, games and puzzles, and everyday task automation, with gaps in current testbeds and guidance for future ones.
Studying how LLMs contribute to planning as solvers, verifiers, and sources of heuristic guidance.
A survey of LLM-assisted planning, covering task decomposition, embodied environments, scheduling, and games.
Improving response generation by learning partial orderings over candidate LLM responses.
Ph.D. in Computer Science · Second year
Advisors: Feng Chen and Bingzhe Li
M.S. in Computer Science
B.S. in Data Science
Bachelor of Economics in Finance
Sep 2026 – Present
Research Collaborator · Persona agents
Persona-aware computer-use agents (PersonaCUABench, co-first author), PersonaScienceBench, and the PersonalAgentEval leaderboard.
Jun 2025 – Present
Graduate Research Assistant · Advisors: Feng Chen and Bingzhe Li · previously Jessica Ouyang
Personal AI, physical AI, and their agentic memory. Earlier, error recovery and multi-agent reinforcement learning for long-horizon computer-use agents, and before that, LLM-based planning and structured reasoning for scientific writing.
Jun 2024 – May 2025
Graduate Research Assistant · Advisor: Sheng Li
Dynamic temporal graphs and language models.
Mar 2023 – May 2025
Graduate Research Assistant · Advisor: Fei Liu
LLM-assisted planning, evaluation, and response ranking.
May 2021 – May 2022
Undergraduate Research Assistant · Advisor: Wei Xu
Feature selection, enterprise graphs, and credit risk propagation.