AI Trading Edge Is Live On Amazon — 25 Chapters On Building A Machine-Learning Trading Stack That Actually Survives Out Of Sample

AI Trading Edge is live on Amazon. Twenty-five chapters, a little under four hundred pages, and a companion code repository you can clone and run.

This is not a memoir and it is not a motivational book. Rogue Alpha is the field manual. The Stock Operator is the story. AI Trading Edge is the build: how you take raw market data and end up with a machine-learning stack that still works on data it has never seen.

Most books in this category either hand-wave the math or hand-wave the market. I wrote the version I wanted when I started wiring models into a real desk — where the data is dirty, the labels leak, and the execution costs eat whatever edge the notebook claimed.

It moves in the order you would actually build it. Data first: ingestion, point-in-time storage, corporate-action handling, and why a survivorship-biased universe will flatter every model you ever fit. Then features: order flow, imbalance, spread and depth dynamics, volatility regimes, and the difference between a feature that describes the past and a feature that carries information forward.

Then the part most people skip — validation. Walk-forward and purged cross-validation, embargo windows, label leakage, multiple-testing inflation, and how to size a claim honestly when you have run four hundred experiments on the same decade of data. If a chapter in this book is worth the cover price, it is that one.

The back half is execution and language models. Reinforcement learning applied to order placement and child-order scheduling rather than to a fantasy of full-position discretion. Then LLM news triage: classification, deduplication, materiality scoring, and the guardrails that stop a headline model from trading a rumor twice.

Every framework in the book is backed by working Python. The companion repository ships production-shaped modules — data loaders, feature transformers, validation harnesses, execution simulators — with unit tests, so you can see exactly what a chapter means instead of reconstructing it from prose.

It is hosted on this site, free, no email gate. Read a chapter, open the matching module, run the tests, then break it on your own data. That last step is the whole point.

Discretionary traders who can already read a tape and want to systematize part of what they do. Engineers who can write clean Python but have never had a P&L punish them for a subtle look-ahead bug. Quants who want a practitioner's account of where research pipelines actually fail on a live desk.

You need comfort with Python and basic statistics. You do not need a PhD, and you do not need a co-location rack. Nothing in the book requires infrastructure a serious retail or small-fund setup cannot stand up.

I have been trading for more than thirty years and building software for traders for most of them — DAS Trader, SpeedTrader, SureTrader. The tooling changed every decade. The failure modes never did: overfitting, ignoring costs, mistaking a regime for an edge, and sizing off a backtest instead of off risk.

Machine learning did not repeal any of that. It industrialized it. A model can now overfit ten thousand ways before lunch, and it will report a beautiful Sharpe while doing it. So the book is as much about discipline and measurement as it is about architecture. The math is there. The skepticism is louder.

Read Rogue Alpha if you want the setups and the risk rules. Read The Stock Operator if you want the thirty-year story — the firms, the trials, the return. Read AI Trading Edge if you want to build the machinery.

They are independent. There is no required order and no series arc. This one is the technical volume, and it stands alone.

No signals, no Discord, no subscription behind this. A book and a repository. If you build something out of it that works, you own it — and if you find a place where I am wrong, tell me and I will fix it in the next edition.

AI Trading Edge is on Amazon today. If it earns it, leave an honest review. That is the only marketing that moves a technical book.