AI Quant Book: A Multi-Agent Curriculum, Not a Strategy Library

What the README Says It Refuses to Be

The README opens with a disclaimer that doubles as a positioning statement: this is not a strategy holy grail. It names three things most quant tutorials stop at, namely API translations of backtesting frameworks, stacked technical indicators with parameter optimization, and overfitted magic strategies. The stated goal is to move a reader from playing at quant to doing quant. That framing defines the audience. The target reader table lists three groups: developers moving into quant, quant researchers who want multi-agent architecture and production risk control, and investors or product managers who want to understand the capability boundary of a quant system. Prerequisites are basic Python, with statistics and market knowledge listed as helpful and machine learning or deep learning explicitly not required. If you already run a live book and want a reference for order routing, this is the wrong shelf.

Five Parts, Twenty-Two Lessons, and Where the Multi-Agent Idea Enters

The content table splits the book into five parts. 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 is only two lessons on supervised learning and the move from models to agents. Part 4 is the largest block at seven lessons, covering architecture, regime detection, LLM use, and risk control. Part 5 closes with five lessons on costs, execution, operations, and projects. The README also states there are 30 background articles and 4 appendices alongside the lessons. The imbalance is deliberate and worth noting: two lessons on machine learning against seven on agent architecture and risk suggests the book treats model choice as the easy part and coordination plus risk as the hard part.

The Data Flow the Book Claims to Teach

The README does not publish a code walkthrough, so the mechanism visible in the material is a division of labour rather than a sequence of function calls. It states that different agents handle different responsibilities, naming signals, risk, and execution, and that these agents collaborate to make trading decisions. The repository references an architecture diagram at manuscript/en/Part1-Quick-Start/assets/multi-agent-architecture.svg, which is the only structural artefact in the README beyond the directory tree. The five questions the book promises to answer map onto that split: where data comes from (rate limits, missing values, adjustments, timezones), how not to fool yourself in backtests (lookahead bias, overfitting, transaction costs), why one model is not enough (regime changes, signal conflicts, risk diversification), how to control risk (stop loss, position sizing, factor exposure, circuit breakers), and how to reach production (execution slippage, monitoring, disaster recovery). Read that list as the syllabus for the agent boundaries.

Getting Started Means Opening a Directory, Not Installing a Package

There is no install command in the README. The quick-start section is two tables of reading paths. English readers are pointed at the manuscript/en/ directory, Chinese readers at manuscript/cn/. The recommended routes differ by background: a complete beginner goes Part 1 then all of Part 2 then Parts 3 through 5; someone with a coding background starts at Lesson 01, skims Part 2, then goes to Parts 3 through 5; someone with a quant background starts at Lesson 01, jumps to Lesson 08, then Parts 3 through 5. The repository tree shows Part1-Quick-Start through Part5-Production under manuscript/en/, plus a Resources-Links folder. The homepage field points to waylandz.com/quant-book-en, which the README presents as the English reading entry. If you were expecting pip install or a requirements.txt, the material gives you neither.

The Licence Is the Sharpest Constraint Here

The work is licensed CC BY-NC-SA 4.0, and the README states this in both languages. That combination has consequences worth stating plainly, without offering legal advice. The NonCommercial term means you cannot fold the text into a paid course or a commercial internal training deck on the terms of that licence. The ShareAlike term means adaptations carry the same licence. The repository metadata supplied here does not list a licence field, and no separate code licence appears in the README, so if lessons include code snippets, the terms covering that code are not established by the material. Anyone planning to lift agent architecture patterns from Part 4 into a proprietary trading stack should confirm the code licensing question before doing so rather than assuming the CC licence settles it.

Where This Book Stops Being the Right Tool

The README is candid that the book is educational and does not constitute investment advice, that strategies are for learning only with no profit guaranteed, and that past performance does not equal future results. Take that at face value and a real limitation follows: a book cannot tell you whether your specific data vendor's adjustment convention matches your backtest, because that depends on the vendor. The five questions are framed as questions, not as solved recipes. The other failure mode is expectation mismatch. A reader who wants a working multi-agent trading system they can point at a broker will find prose about agent responsibilities, regime detection and circuit breakers, and will still have to write the signal, risk and execution layers themselves. The repository structure shows only manuscript directories and a README, so there is no engine to inherit.

How It Differs from a Backtesting Framework Tutorial

The natural comparison is the category the README criticises: a tutorial built around one backtesting framework's API. That kind of material is executable and immediately rewarding. You copy a snippet, feed it a CSV, get an equity curve, and the framework handles the event loop, the order model and the reporting. ai-quant-book inverts the trade. It gives up executability in exchange for coverage of the parts a framework hides from you: rate limits and timezone handling on the data side, lookahead bias and transaction costs on the validation side, and the coordination problem of reconciling a signal agent with a risk agent that may disagree. A framework tutorial teaches you one tool's vocabulary. This book, by its own description, teaches the questions you should be asking before you pick the tool. Neither approach substitutes for the other, and the reading paths in the README assume you have not yet chosen.

Maintenance Cost and What to Verify Before Committing

The repository is not archived and the last push recorded in the supplied metadata is 2026-08-08. No releases were retrieved, which is consistent with a book repository that ships text rather than versioned artefacts. The practical maintenance question is different from a library's. There is no dependency graph to update and no breaking API to track, but there is a content decay risk: anything the book says about LLM behaviour in Part 4, or about specific exchange rate limits in Part 2, ages at the speed of those systems rather than at the speed of the prose. The README claims both language versions are complete at 22 lessons, 30 background articles and 4 appendices, and the reading paths assume that structure. Check that claim against the actual files in manuscript/en/ and manuscript/cn/ before you plan a study schedule around it, and check whether the licence file exists in the repository or only as a badge and a line of text.

Editorial conclusion

Adopt it if you write Python and want a structured reading path through data plumbing, backtest bias and multi-agent risk design before you commit to a framework. Skip it if you need an installable library, an API reference, or a backtest engine you can run today; the repository ships prose, not code you execute. Verify two things first: whether the repository carries any licence file or code samples beyond the CC BY-NC-SA 4.0 text licence, and whether the English manuscript directories under manuscript/en/ are as complete as the README table claims.

Official sources

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