Software Engineering student · Python · Desktop apps · Automation · Local LLM applications
| Project | Description | Highlights | Stack |
|---|---|---|---|
| private-cinema | Local media library manager — scan, parse, poster, resume playback, stats | 438 tests · 5.1k LOC · CI on Win × py3.10–3.12 · 732 GB library verified | PySide6, SQLAlchemy, ffmpeg, mpv |
| liwangtong | Campus network auto-login — tray daemon, auto-reconnect, secure storage | 3-process design · named-mutex single instance · Windows Credential Manager · MIT | Python, ctypes, pystray, Tkinter |
- Testability as a design constraint — e.g. pluggable embedding layer so the whole retrieval pipeline is testable without a GPU, a model download, or network access.
- Failure should be loud — mixing vectors from two different models must raise, not silently return meaningless similarity scores.
- Understand every line — I use AI-assisted coding, but review, debug and take responsibility for each decision in my projects.
- 2 public projects, both MIT-licensed and released
- 438 pytest cases on
private-cinema, run in CI on Windows × Python 3.10–3.12
📍 Huainan, China