M.S. Candidate @ Tianjin University · Wave–Structure Interaction · Research Software · Reliable AI Tooling
Research · Systems · Open Source · Signals · Contact
I study computational coastal engineering and build scientific software and reliable AI systems. My bias is toward explicit assumptions, reproducible evidence, and systems that remain correct when things go wrong.
| 🎓 Tianjin University M.S. Candidate |
🌊 Research Wave–Structure Interaction |
⚙️ Compute LBM · GPU/HPC |
🤖 Systems Agents · MCP · Automation |
My research centers on waves interacting with coastal and porous structures, with an emphasis on physically defensible numerical results rather than tool loyalty.
| Scientific questions | Computational methods |
|---|---|
| Wave–structure interaction · free-surface flow · multiphase flow · porous-media interaction | Lattice Boltzmann methods · GPU/HPC workflows · numerical diagnostics · verification & validation |
Current principle: the scientific question owns the route; solvers, frameworks, and agents are replaceable means for reaching a result that can be checked and reproduced.
Public research artifacts will be linked here when they are documented, reproducible, and cleared for release.
Switch agents. Keep your context. A local-first, user-owned context layer that lets independent AI agents share durable context without collapsing into one role.
What it demonstrates: cross-agent continuity, MCP integration, transactional writes and rollback, deterministic validation, explicit authority boundaries, and an isolated Claude → Codex handoff proof.
🎓 Tutor
A file-based AI tutoring template built around persistent learner state, retrieval-oriented sessions, review scheduling, and reusable agent instructions.
Engineering focus: stateful workflows, inspectable file-based state, configurable learning gates, and multi-agent project structure. Educational-effect claims are kept separate from software behavior.
I prefer contribution work that starts from a reproducible failure and ends with a narrow fix, regression coverage, and a clear upstream decision.
| Upstream project | Contribution | Status |
|---|---|---|
| NousResearch/hermes-agent | Prevented destructive context compression after degraded empty provider responses; substantive fix from #94531 was incorporated through #94606 with authorship preserved. | ✅ Incorporated upstream |
| Draculabo/AntigravityManager | Fixed 401 token recovery, concurrent poll coordination, and transient availability-state cleanup with regression coverage in #305. | ✅ Merged upstream |
Ongoing upstream contributions
- OpenClaw #138750 — preserving the dotenv assignment backing a newly configured environment SecretRef. Open / not merged as last checked 2026-09-05.
- CC Switch #6791 — routing takeover for declared long-context model markers. Open / not merged as last checked 2026-09-05.
reproduce → isolate → fix → test → review → upstream
Working principle: reproduce the failure, make assumptions explicit, test the fix, preserve the evidence.
I use AI throughout development and research workflows; problem definition, validation, and technical judgment remain my responsibility.
Additional experience: WorldQuant BRAIN Consultant.
I welcome conversations about research collaboration, scientific computing, research software, and reliable AI systems.
