Data engineer. Nearly three years building the market-data platform behind a Shanghai quant asset manager, then independent consulting for private quant clients. Moving to the Netherlands and looking for sponsored data engineering work.
Most of what I write is about the unglamorous half of data: getting it in, knowing it is correct, and knowing when it is late. A wrong number before the open is a trading loss, not a bad chart.
| Repo | What it is |
|---|---|
| data-quality-gate | Stops a daily pipeline when a symbol goes stale, a day disappears, a column is renamed or rows duplicate. Exits non-zero so CI gates on it. |
| tick-rollup | Streams a trade CSV into OHLCV bars, with hard session boundaries and two documented late-tick modes. Measured at ~260k ticks/s on this machine. |
| rag_im | Snapshots government visa pages, diffs them section by section, answers questions with citations. Local-first, no API key. |
| mahjong-mc | Plays millions of Japanese Mahjong hands to compare strategies. Pure Python, no dependencies, 114 tests. |
| GapFiller | Monte Carlo Yahtzee: hundreds of thousands of games per strategy, reported with standard errors instead of opinions. |
| PacePulse | Offline-first HIIT and strength timer, shipped as a web app and an Android wrapper. |
Python, SQL and Bash; Spark, PySpark, Hive and HDFS; Airflow for orchestration; pandas and NumPy for analysis; dbt (certified, and putting it to work); LlamaIndex and ChromaDB for retrieval work. HDF5 and Parquet are where the data actually lives.
Email: louis.xiaxu.tian@gmail.com · LinkedIn

