Jiahao Zhang

I’m Jiahao Zhang (张嘉浩), a PhD student in Computer Science at Vanderbilt University, advised by Prof. Yu Huang. I build and evaluate AI coding agents and large language models for software engineering, with a focus on reliable code repair, planning, memory, and efficient decision-making.

My recent work includes:

• JevCoder: Delegating repair decisions to a low-cost probabilistic judge. Across three repair models, each evaluated on the same 400 tasks, it reduced mean repair-model token use by 21.6%–50.6% versus unguided repair, with higher observed repair rates.
• PMCoder: Coupling hierarchical planning with episodic memory so that task progress guides retrieval and memory feedback triggers replanning during software issue resolution.
• Human attention for code LLMs: Incorporating eye-tracking signals into model training and code comprehension through EyeLayer (ICPC 2026) and EyeMulator (ACL 2026).

I work with Python, PyTorch, Hugging Face, SFT/DPO, vLLM, and SWE-bench evaluation harnesses. My experience spans agent workflows, model training, and empirical evaluation.

Previously, I earned my bachelor’s degree at Southern University of Science and Technology (SUSTech) with Highest Honors in Computer Science and Engineering, supervised by Prof. Yepang Liu. I also worked with Prof. Joanna C. S. Santos as a research intern at the University of Notre Dame.

CV · GitHub · Google Scholar · Email