Production-Grade Data Science book cover
Book in progress · coming December 2026

Production-Grade Data Science

An engineering handbook for working data scientists.

You can train a model. The question this book asks is harder: can you prove it works, ship it, and know the moment it breaks?

Most data science code never clears the bar. It runs once, in one notebook, on one machine, and quietly fails everywhere else. The gap isn't statistics or modeling. It's engineering: reproducibility, testing, packaging, deployment, and monitoring: the discipline that separates a script that ran from a system you can stand behind.

This book is that standard, written by a software engineer who builds production systems and teaches data scientists to do the same. It's organized as six engineering standards every production data scientist should meet, with reliability as the line you can't cross. Every chapter ends not with a summary but with a verdict you can apply to your own work.

The six standards

0

Reproducible

Can someone else run it and get your result?

1

Legible

Can a stranger (or you at 3 a.m.) read it?

2

Structured

Is it a system, or a pile of scripts?

3

Proven

Is there evidence it works, not a vibe?

4

Shipped

Can it run where the users are?

5

Accountable

Would you know the moment it breaks?

Read the introduction

The opening chapter, "The Passing Grade": the whole argument in five pages.