ai-quant-book teaches system building and ships no code

The repository is a manuscript directory

The whole tree is a README, a .gitignore and the manuscript. Both language versions are marked complete, each with 22 lessons, 30 background articles and four appendices:

ai-quant-book/
├── manuscript/
│   ├── cn/                      # 中文版 (Complete 已完成)
│   │   ├── Part1-快速体验/
│   │   ├── Part2-量化基础/
│   │   ├── Part3-机器学习/
│   │   ├── Part4-多智能体/
│   │   └── Part5-生产与实战/
│   ├── en/                      # English (Complete)
│   │   ├── Part1-Quick-Start/
│   │   ├── Part2-Quant-Fundamentals/
│   │   ├── Part3-Machine-Learning/
│   │   ├── Part4-Multi-Agent/
│   │   └── Part5-Production/
└── README.md

Each language directory ends with a Resources and Links folder, which is where the 30 background articles live. There is no `src` directory, no package manifest, no notebooks and no tests, and the English edition is also hosted away from GitHub on the author's own site. That matters for what the book claims to teach, since its whole argument is that building production systems is the hard part and tutorials usually stop before it.

Lesson counts put the weight where the author wants it

Five parts, and the distribution is the argument in numbers. Part 1 is a single quick-start lesson covering the quant landscape and multi-agent intuition. Part 2 spends seven lessons on fundamentals: markets, statistics, strategies, data and backtesting. Part 3 gets two lessons on machine learning, moving from supervised learning to agents. Part 4 returns to seven lessons for the multi-agent material, covering architecture, regime detection, LLM use and risk control, which is where regime detection, the idea that market behaviour changes and one model cannot cover every phase, gets its own treatment. Part 5 closes with five lessons on production: costs, execution, operations and projects. Two lessons of machine learning against seven of multi-agent and five of production tells you the book is not trying to teach model fitting.

The Core Architecture heading has an empty body

Between the content overview and the target readers there is a section titled Core Architecture, and under that heading there is nothing but a horizontal rule, which is odd since it is the one section a reader cannot reconstruct from the lesson titles. The architecture is described elsewhere, in the overview, in one sentence: different agents handle different responsibilities such as signals, risk and execution, and they collaborate to make trading decisions. So the diagram every multi-agent book needs is the one artefact missing from the repository. A reader who wants to see how a signal agent talks to a risk agent has nothing to look at and no code to read, only prose describing that they collaborate.

Five questions, and an argument against strategy recipes

The premise is stated as a refusal: not a strategy holy grail, but teaching you to build production-ready quant systems. What it criticises in other tutorials is API translations of backtesting frameworks, stacked technical indicators with parameter optimisation, and overfitted magic strategies, on the grounds that these let you play quant rather than do it. Each of those five questions is asked twice in the README, once in English and once in Chinese, and the answer is meant to be the same architecture in both languages. The replacement for the recipe is a list of questions. Where does data come from, with rate limits, missing values, adjustments and timezones. How not to fool yourself in backtests, naming lookahead bias, overfitting and transaction costs. Why one model is not enough, because regimes change, signals conflict and risk wants diversifying. How to control risk through stop loss, position sizing, factor exposure and circuit breakers. And how to reach production, covering execution slippage, monitoring and disaster recovery.

Three reading paths, none of them from zero

The quick start offers a path per reader type, and all three assume something. A complete beginner goes Part 1, then all of Part 2, then Parts 3 to 5. Someone with a coding background starts at lesson 01, skims Part 2 and continues. Someone with quant background starts at lesson 01, jumps to lesson 08, then reads Parts 3 to 5, which is a shortcut that assumes you can skip seven fundamentals lessons. Prerequisites are stated plainly: basic Python programming is required, statistics and financial markets help, and a machine learning or deep learning background is explicitly not needed. Target readers are developers moving into quant, quant researchers who want the multi-agent and production risk material, and investors who want to understand what such a system can and cannot do.

CC BY-NC-SA 4.0 in the README, no LICENSE file in the tree

The licence story needs reading carefully. The README states the work is licensed under CC BY-NC-SA 4.0 and shows the Creative Commons badge, in both language sections, while the repository's own licence metadata does not resolve to a recognised identifier and no LICENSE file appears among the top-level files. Attribution, non-commercial and share-alike terms therefore live in prose rather than in a file a tool can read. For a text work that is mostly a formality; for anyone who wants to reuse the material inside a company training programme, the non-commercial clause is the part to check first, because a course sold to a paying audience is not obviously covered by it.

No releases, one author, and a risk disclaimer in the README

Maintenance signals are thin. The repository publishes no GitHub releases, so there is no version to cite and no changelog to diff, and the only attribution is to Wayland Zhang, who also links a hosted English edition of the book away from GitHub. The last push to the default branch was on 2026-08-08. In the repository's favour, the README front-loads the disclaimer rather than hiding it: quantitative trading involves risk, the book is educational and not investment advice, strategies are for learning with no profit guaranteed, you should understand the risks before live trading, and past performance says nothing about future results. It closes with two epigraphs, one from Jesse Livermore on the stock market repeating itself and one from Jim Simons about searching history for patterns believed to be predictive of future price action. Taken together they set the tone the rest of the disclaimer confirms: this is material about how to build and test systems, not a promise about what markets will do.

Editorial conclusion

ai-quant-book fits a developer who already writes Python and wants a structured map of what stands between a backtest and a live system, in either language, at no cost. It does not fit someone looking for runnable agents, a strategy to trade, or depth on machine learning, since Part 3 is two lessons and the repository contains no code. Before you start, read the risk disclaimer the book puts in the README, and check the lesson count against the two lessons on machine learning so you know where the emphasis is not.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. waylandzhang/ai-quant-book on GitHub