A local quant workbench for China’s stock market tinkerers

What it does

TickFlow Stock Panel is a web-based quantitative workbench built specifically for China’s A-share market. It packs screening, real-time monitoring, and backtesting into one deployable unit, keeping K-line, financial, and alert data in local Parquet and DuckDB instead of a vendor’s cloud. The stack is React on the frontend and Python with Polars handling vectorized scans on the backend.

The interesting bit

Strategy development works like a plugin system: write Polars expressions by hand, generate them from a sentence via an LLM (with ast validation), or import CSVs and JSON from other quant projects into DuckDB for side-by-side analysis. The README also explicitly swears off “AI stock picking,” which is a rare moment of honesty in a space crowded with miracle-cure marketing.

Key highlights

• 20 built-in screening strategies covering trend, volume/price, and reversal patterns, all vectorized via Polars.
• Optional LLM strategy generation through any OpenAI-compatible API; leave the config blank and the feature simply disappears.
• Backtesting enforces A-share realities: T+1 settlement, fees, slippage, stop-loss, and maximum holding periods.
• Third-party data ingestion with automatic schema discovery, so your own datasets can sit next to TickFlow data in DuckDB.
• A post-market pipeline that automatically pulls daily K-lines and recalculates enriched indicators at 15:30 CST.

Caveats

• Tushare integration is still marked as in development, and the individual stock analysis module is explicitly labeled Beta.
• Backtest dependencies rely on vectorbt and numba, which may require manual compilation on macOS Intel hardware.
• Older CPUs missing AVX2/FMA instructions need a legacy-compatibility flag for Polars to start.

Verdict

A solid fit for retail traders and quant hobbyists who want to tinker with A-shares without shipping data to a black-box terminal. If you trade outside China or need fully baked institutional tooling, this isn’t your stack.