Undergraduate at the School of Artificial Intelligence, Shanghai Jiao Tong University, building AI systems that can be tested, verified, and reproduced.
- 🎓 Undergraduate (Junior) at SJTU School of Artificial Intelligence
- 🎓 Research Intern at Microsoft Research Asia (MSRA) in Shanghai
- 🔍 Interested in Self evolving agents & RSI, reliable agents, repository understanding, and scientific AI
- 🧪 I care the definition of the capability of an agent and about the gap between "the agent claims success" and "the task actually works"
- 🏅 National Scholarship recipient (Top 0.5%, 2025)
- 🚀 Admitted to university one year early through Gaokao (Top 0.05% in Jiangsu)
An LLM-powered multi-agent system that configures software repositories inside Docker using reusable XPU knowledge, speculative execution with rollback, and independent adversarial verification.
A reproducible, budget-aware framework for self-supervised speech recognition. It combines fast and strong SSL recognizers with complementarity-aware routing and representation-compression analysis.
|
A reproducible comparison of StreamingLLM, SnapKV, and PyramidKV on Pythia-70M, covering perplexity, memory usage, latency, and long-context trade-offs.
|
|
A local web application for blind and model-assisted mass-spectrometry structure annotation, with per-peak evidence, progress tracking, and exportable results.
|
LLM-assisted repository reproduction with reusable execution experience, knowledge retrieval, parallel processing, and containerized validation.
|
- Reliable LLM Agents & Tool Learning — long-horizon decision making, self-verification, and failure-aware control
- Code & Repository Understanding — automatically configuring and reproducing real-world software repositories
- Agent Evaluation — execution-based benchmarks and adversarial checks that separate claims from verified outcomes
- Scientific AI — combining structured data, code, and natural language for scientific workflows
