AI undergraduate · Product-minded builder · Reproducible systems
I turn ambitious ideas into runnable products—from LLM applications and 5G digital twins
to local-first developer tools with real installers, tests, and recovery paths.
- AI products that connect language models with real data, maps, workflows, and user constraints.
- Intelligent systems with simulation, surrogate models, uncertainty-aware optimization, and closed-loop evaluation.
- Developer tools that feel native, stay local-first, and remain easy to install, verify, and restore.
|
An end-to-end AI travel companion for weekend group trips: intent understanding, real POI and routes, budget-aware planning, live replanning, and an interactive companion. Evidence: 314 pytest cases · GitHub Actions CI · user-supplied LLM and AMap keys · multi-platform launch scripts
|
A 5G digital-twin platform for closed-loop A3 parameter optimization, combining radio simulation, ONNX surrogate models, risk-aware search, and an interactive dashboard. Evidence: reproducible experiment scripts · uncertainty-aware optimization · architecture contract · validation report
|
|
A lightweight Codex usage meter and quick-control dock with responsive native-style UI, focus controls, local session accounting, and one-click installation. Evidence: Windows and macOS installers · automated tests · release packages · one-click restore
|
Codex Dream Skin forkOpen-source work on cross-platform Codex theming through local CDP injection—custom backgrounds, saved themes, native controls, and safe restoration without modifying the official package. Focus: Windows and macOS tooling · visual customization · local safety boundaries · contributor workflows
|
Make it runnable. Measure the result. Keep the boundary explicit. Leave a recovery path.
Building in public, one validated system at a time.