B.Eng. in Software Engineering @ Tongji University
Incoming M.S. Student @ Shanghai Jiao Tong University (2027)
Exploring the intersection of Software Security, Large Language Models, and Software Engineering.
🎓 I'm currently an undergraduate student in Software Engineering at Tongji University. I have received pre-admission to a master's program at Shanghai Jiao Tong University (SJTU) and expect to begin my graduate studies in 2027.
🔐 As I prepare for graduate study at SJTU, I look forward to further exploring Software Security, LLM / Agent Security, and Automated Vulnerability Repair.
🧠 I'm particularly interested in combining large language models and program analysis to understand, detect, and repair software vulnerabilities.
🚀 Currently focusing on LLM-based program repair, complex vulnerability fixes, context engineering, and secure AI agents.
💡 I enjoy building practical systems at the intersection of AI × Security × Software Engineering.
🥁 Outside of research, I enjoy drumming and 🎾 tennis.
Important
LLM-based Automated Vulnerability Repair
My primary research focuses on leveraging large language models to repair software vulnerabilities, especially complex vulnerabilities that require richer contextual information, multi-step reasoning, and coordinated modifications across multiple program locations.
🧩 Multi-hunk Vulnerability Repair
Repairing complex vulnerabilities whose patches span multiple code locations.
🧠 Context Selection & Organization
Identifying useful program context and presenting it effectively to large language models.
🔄 Repair Workflow Design
Designing structured multi-step workflows for complex vulnerability repair.
🔍 Program Analysis + LLMs
Combining static/program analysis signals with the reasoning capabilities of foundation models.
✅ Reliable Security Patching
Improving the correctness and reliability of automatically generated vulnerability patches.
Note
MCP Ecosystem Security
Studying security risks in the Model Context Protocol ecosystem, with a focus on vulnerable information flows and missing security information in MCP servers.
Exploring metadata-enhanced and runtime defense mechanisms for securing LLM agents.
Tip
Dormant Ransomware Detection
Exploring techniques for detecting dormant ransomware by combining program analysis, LLM-derived semantic information, and graph-based learning.
Focusing on identifying malicious behaviors that remain hidden before activation.
I'm always interested in discussions and collaborations around
Automated Vulnerability Repair · Software Security · LLM / Agent Security · AI for Software Engineering
Software Security · LLM Security · AI for Software Engineering


