Building intelligent systems, from first principles.

An essay on composition as the real unit of AI engineering. Capabilities change; patterns of assembly endure. Free PDF, no email required.

Three parts: the nature of composition, the transformations (sound, documents, images, text, video), and the practice. Not a tutorial. A way of thinking that survives model generations.

§ 01 · What’s inside

Three parts. One arc.

Part I

The Nature of Composition

The history and philosophy of AI transformations. Why modularity wins. The bitter lesson and the economics of scale.

Part II

The Transformations

From sound, documents, images, text, and video. Each modality and the hinges that connect them.

Part III

The Practice

Patterns of assembly. Evidence-based prompting. The craft of building systems that actually work.

§ 02

Why this book

Compose intelligence, don’t chase models.

“The goal is not to create intelligent machines, but to compose intelligence from simple transformations.”

This is not another AI tutorial with code snippets that will be outdated next month. It is a way of thinking about intelligent systems that stays useful regardless of which models dominate tomorrow.

No hype. No specific-model worship. No LinkedIn frameworks — just research-backed techniques and a few hard-won opinions about where composition really pays off.

§ 03 · Table of contents

Every chapter, listed.

Part I

The Nature of Composition

Prologue · A brief history of transformations
On the art of assembly
Principles of composition
The bitter lesson
The economics of scale

Part II

The Transformations

From sound
From documents
From images
From text
From video

Part III

The Practice

Patterns of assembly
On prompting
The craft of composition

Appendices

Reference

Transformation taxonomy
Implementation notes

§ 04 · About the author

Kacper Wikiel.

Builds AI systems at the intersection of research and production. Runs Codesota, documenting what actually works in applied machine learning.

This book distils years of building, breaking and rebuilding AI systems into principles that transfer across domains and survive model generations.

Start reading

Free PDF. No email. Anonymous counter only.