A Chinese quant kitchen sink: every AI trick thrown at the market

What it does This repo is a sprawling, un-packaged knowledge base for Chinese retail and institutional traders who want to apply every modern AI technique to financial markets. It bundles Jupyter notebooks, example strategies, and resource lists covering stocks, funds, and crypto across local backtests, online platforms (JoinQuant), and simulated live trading via Wind data terminals.

The interesting bit The sheer breadth is the point. The author has assembled what they claim is a “史上最全” (most complete ever) collection of quant tools, annotated with personal reviews and gotchas—less a framework, more a curated flea market of techniques from tsfresh auto-factor-mining to Unsloth-tuned LLMs for “interpretable” price prediction.

Key highlights

• Strategy zoo: vanilla rules, ML, deep learning, reinforcement learning (SB3/FinRL), graph networks, and high-frequency snippets
• LLM frontier: includes a May 2025 Unsloth stock forecaster and “reasoning-type” price-prediction model training tutorials
• Deployment spectrum: Python/C++, CPU/GPU, with a runtime/ directory for actual execution
• Data plumbing: Wind terminal integration, open-source data tools, and an Excel-based “market-watching artifact”
• Resource curation: actively maintained notes, paper summaries, and a 26k-view Zhihu article indexing the quant open-source landscape

Caveats

• Not a library: explicitly “not yet packaged as a Python package”; copy the whole repo and hunt through egs_* directories
• Freemium model: some newer code is “privatized” behind a paid knowledge-planet community; the README is peppered with QR codes and discount offers
• Claims like “prediction accuracy improved 20%” and “annualized return 53%” appear without reproducible benchmarks or Sharpe ratios attached

Verdict Worth bookmarking if you’re a Mandarin-speaking quant hobbyist who wants to survey the full AI-for-trading landscape without building from scratch. Skip it if you need a maintained, testable framework with clean APIs—this is a learning scrapbook, not a trading system.