I work on the infrastructure other code trusts — linters, language runtimes, and locale data. A wrong answer there is silent and expensive.
- 2026 – M.Sc. in machine learning and big data, CCNU–UOW Joint Institute
- 2024 – 2026 Trading systems for A-share, HK, US and futures markets: CTP, Futu and QMT gateways, a C++ SDK wrapped into CPython, an Electron strategy terminal
- 2024 Access control and behaviour tracking for hospital operating rooms: face and IC-card identity, ISO 9001 / CE acceptance
- 2020 – 2024 B.Eng. in computer science and technology, Inner Mongolia University of Technology
| Project | Change |
|---|---|
| pylint-dev/pylint #11650 | pylint --source-root . pkg/mod crashed when the package was an implicit namespace package that also exists in site-packages — merged, milestone 4.1.4 |
| pylint-dev/pylint #11703 | --load-plugins checkers emitted every message twice under --jobs > 1 — merged, backported to 4.1.x |
| pylint-dev/astroid #3445 | The astroid side of #11650: is_namespace() can take a search path, so a namespace package shadowed by site-packages is still recognized — in review |
| PyCQA/isort #2728 | An action comment on a continuation line ends the sort section only on the next run — changes requested; conflicts with upstream |
| python/cpython #160664 | test_subprocess.test_getoutput decoded cmd.exe's OEM output as UTF-8, so it failed on every non-English Windows — approved |
| python/cpython #160716 | The remote PDB interrupt listener half-closed its socket, so Windows dropped the channel after 120 idle seconds — approved |
| python-babel/babel #1383 | CLDR 49 ordinal day-of-month patterns for ddd — approved; waiting for CI and merge |
| ruby/TryRuby #258 | Typos and dropped words in the Chinese translation, lessons 1 to 20 — in review |
plum — A-share backtesting and live trading where the market's rules live in the engine, not in the strategy. Board-level price limits (20% on ChiNext and STAR, 5% on ST), T+1 settlement, halts, delisting settlement at the last close, square-root market impact capped by participation, point-in-time fundamentals. Cash and the ledger use Decimal, so the book never drifts on float error.
kt-audit — A pip-installable audit for knowledge-tracing experiments. It answers what the protocol usually leaves open: whether train events precede test events for each user, whether items leak across splits, and whether the reported AUC matches a recomputation from the released predictions. In the pyKT benchmark library, data_process.py points valid and test at the same file, and its own authors warn the results cannot be compared. kt-audit exists so that check runs before anyone cites the number.
| Layer | Used |
|---|---|
| Backend | Python (Flask, FastAPI) · Go (Gin) · Ruby (Rails) · C++ |
| Frontend & desktop | React · Vue 3 · Next.js · Electron · PyQt6 |
| Infra & data | Docker · nginx · PostgreSQL · Redis · MongoDB · DuckDB · DolphinDB |
| Daily drivers | uv · pnpm · ruff · biome · pytest · vitest · gh |
Reproductions come first; a fix without a failing test is a guess.