I am an M.S. student in Software Engineering at Shenzhen University. Previously, I was a research intern at Ant Group's Technology Research Institute. I build LLM agents for real-world code repositories, with a focus on context engineering, tool use, multi-agent orchestration, and reliable execution. My current work applies these ideas to Rust code generation and verification in the Asterinas ecosystem.
I am interested in how agents turn a task into a working change: retrieving relevant context, planning edits, calling tools, learning from execution feedback, and checking the result. My work spans both agent workflow engineering and research on repair decisions and output quality. I am also interested in evaluating the quality of agent-generated code and formal specifications, including code correctness, specification completeness, and alignment with intended behavior.
- Context engineering — Retrieve repository history, cross-file dependencies, and reusable proof patterns; load domain knowledge on demand through shared skills.
- Tool use & orchestration — Compose generation, migration, verification, repair, and review into reusable workflows, with explicit tool constraints and retry budgets.
- Multi-agent collaboration — Separate planning, specialized repair, patch execution, and structural rewriting so each role handles a distinct decision in the repair loop.
- Evaluation & auditing — Examine verification results, specification quality, and executable semantics separately to assess what an agent's output actually guarantees.
- KVerus Skills — Agent workflow engineering during my internship at Ant Group. Contributed reusable skills that connect repository context retrieval, Rust-to-Verus migration, proof generation, and iterative repair through verifier feedback. The workflow also incorporates post-processing review gates and semantic auditing for Asterinas kernel verification.
- StarVerus — Multi-agent generation and repair. First-author work accepted to SIGKDD 2026. Builds repository context from caller/callee relationships, aligns generated contracts, and coordinates Planner, Repairer, Actor, and Rewriter roles within a bounded verification-and-repair loop.
- Beyond Benchmarks — Understanding agent failures in real repositories. Third-author work accepted to ASE 2026. Studies how repository dependencies and semantic modeling affect coding agents, why verifier diagnostics can diverge from root causes, and how iterative repair can stagnate or regress.
- Large Language Models for Multilingual Code Intelligence: A Survey — First-author survey accepted to KSEM 2026, covering multilingual code generation, translation, retrieval-augmented generation, multi-agent methods, and trustworthy evaluation.
Coding Agents · Context Engineering · Tool Use · Multi-Agent Systems · Agent Skills · Feedback-Driven Repair · Agent Evaluation · Code & Specification Quality · RAG
Python · Rust · LangGraph · Claude Code · Codex · Function Calling · Verus · Static Analysis
- M.S. in Software Engineering, Shenzhen University, 2024–2027
- B.Eng. in Software Engineering, Changsha University of Science & Technology, 2020–2024
- GitHub: Je5s1e
- Email: chao_cc@outlook.com