At Huawei, I developed a multi-agent LLM system for automated merge-request code review, with workflow orchestration, tool calling, repository analysis, and review-result aggregation. I also designed a developer-feedback mechanism to support iterative improvement of review quality.
I built an AI-assisted development analytics platform that combined AI-generated code metadata with repository and merge-request data. Its data services and dashboards analyzed AI-generated code ratios, repository activity, and development metrics across teams.
At HKUST(GZ), I researched cross-chain transaction information leakage and MEV sandwich attacks. I analyzed blockchain and protocol transaction data, developed methods to detect suspicious trading patterns, and contributed to a paper on attacks that generated over USD 5.27 million in profit in an empirical study of Symbiosis.
In my multi-document RAG project, I developed a pipeline that selects complementary evidence and checks context sufficiency before answering or abstaining. Set-wise evidence selection improved all-evidence recall from 3% to 14% under matched passage budgets while maintaining valid citations.

