About
I am an incoming Ph.D. student at The Hong Kong University of Science and Technology (Guangzhou), where I will join the MAYU Lab under the supervision of Prof. Zhensu Sun.
Previously, I was a Research Assistant at the LLMSE Lab, Shanghai Jiao Tong University, from February 2025 to December 2026, supervised by Prof. Xiaodong Gu and Prof. Beijun Shen.
My research focuses on Large Language Models for Software Engineering (LLM4SE), particularly developing more capable, autonomous, and reliable coding agents.
My research interests include:
- Software Engineering Agents — improving the capabilities of coding agents in complex, real-world software development tasks.
- Agent Training and Self-Improvement — exploring effective training strategies and synthetic data generation for coding agents.
- Software Engineering Benchmarks — designing realistic and challenging benchmarks to evaluate the capabilities and limitations of coding agents.
- Repository-Level Code Generation — enabling autonomous agents to generate, maintain, and evolve software repositories.
News
Our paper RSIHub: A Composable Framework for Agent-Evolution Methods has been accepted to Meta-Agents Workshop at NeurIPS 2026.
Our paper Context Compression for LLM Agents has been accepted to EMNLP 2026.
Our paper Planning over Actions: Agentic Reasoning for Semi-Structured Table Question Answering has been published online in Information Processing & Management.
Our paper Test vs Mutant: Adversarial LLM Agents for Robust Unit Test Generation has been accepted to ISSTA 2026.
Our paper LLM Agents Can See Code Repositories has been accepted to ASE 2026.
Our paper LLMs as Continuous Learners: Improving Software Issue Reproduction by Distilling Experiences has been accepted to Internetware 2026.
SWE-Exp ranked 2nd on the SWE-bench Verified leaderboard among methods using both open-source models and open-source scaffolds!
Publications
Google Scholar ↗Newest first · publication or first release dates; venue labels show conference years.
2026
SWE-at-Scale: Scalable Issue Synthesis from Historical Bug Fixes
LLM Agents Can See Code Repositories
Test vs Mutant: Adversarial LLM Agents for Robust Unit Test Generation
Schrödinger’s Code Repository: Have LLMs Learned SWE-bench or Memorized It?
Repo0: Design-Driven Zero-to-All Code Generation
SkillForge: Self-Distilling Agents for Project-Specific Issue Resolution
Planning over Actions: Agentic Reasoning for Semi-Structured Table Question Answering
Know Before Fix: QA-Driven Repository Knowledge Acquisition for Software Issue Resolution
Context Compression for LLM Agents: A Survey of Methods, Failure Modes, and Evaluation
2025
SWE-Exp: Experience-Driven Software Issue Resolution
2024
LLMs as Continuous Learners: Improving Software Issue Reproduction by Distilling Experiences
No works match these filters.