Introduction

docs.scrimba.com

Juno

Welcome aboard Crooked Orbit! I'm Juno, and I'm so glad you're here. Getting new engineers comfortable with the AI systems we run is my favourite part of the job, and watching this stuff go from magic to something you actually understand never gets old.

We'll start with what these models really are under the hood, no machine learning background needed, and build it up one piece at a time.

Nothing here is ever too basic to ask about, and trust me, it took me way longer than I'll admit to get some of it. Take it at your own pace and we'll get there together!

Juno

Welcome aboard Crooked Orbit. I'm Ace, systems engineer, and I keep the day-to-day AI builds shipping around here. You've got the basics, so we'll move at a steady pace and stay practical, the way you'd actually wire this up in real code.

Every chapter is the concept, then what to do with it: read it, try it, build it. Half the systems on this station run on exactly that loop, so you're in good company.

Juno

Welcome aboard Crooked Orbit. I'm Sol, chief engineer, fourteen years and too much coffee aboard this place. These are the docs for people who want to know how the AI systems really work, not only which button to press.

I've wired most of this myself, usually twice, because the first version was a mistake. Expect the trade-offs, the failure modes, and the details everyone else skips. Read closely and question all of it, including me.

This handbook is for developers who want to understand how AI actually works, not only call an API and hope. A quick word on two terms you'll see throughout: a model is the AI system you send text to and get text back from, like the one behind ChatGPT, and a prompt is the text and instructions you send it. Calling a model is an API call: your code sends it your prompt over the web and gets text back, the same as any web service. The concepts here are written out in full, with depth, examples, and something to search whenever you need a quick answer.

These docs assume you can already write Python: variables, functions, and calling a web API. If that sounds like you, you have everything you need; if you want to pick up the basics first, the Python handbook covers them from scratch. You do not need any machine learning background. AI engineering is building products with models other people trained, not training your own.

The approach here is to explain the machine, not only the API. Anyone can copy a model call from a tutorial and get something working on the happy path. The distance between that and shipping something reliable is understanding what happens underneath: why the model produces what it produces, why it fails in the particular ways it does, and what you can do about each one.

So this handbook spends real time on how these models work before it shows you what to type. It is written for the developer who would rather understand the engine than only turn the key, because understanding the machine is what outlasts any single model. That is what lets you debug a wrong answer instead of shrugging at it, whichever model you reach for next year.

These docs teach the concepts in depth, and you'll put them into practice here with exercises and projects of your own. Scrimba's main course is the guided, build-along version of the same material.

Take the AI Engineer Path on ScrimbaBuild these concepts into real apps, hands-on in JavaScript→

Python here, JavaScript in the course

These docs are written in Python; the AI Engineer Path teaches the same material in JavaScript for the web. That difference matters less than it looks: AI engineering is the same work in any language, where you send text to a model over a web request and shape what comes back. Every idea here, from tokens and prompting to retrieval, agents, and evals, is identical whatever you write it in; only the syntax around the model call changes. Learn it in one and you already know it in the other, so reading here and building in the course reinforce each other.

JunoA fast-moving field The model names and prices in this field change almost monthly, so don't worry about memorising them. We focus on the parts that hold still: how these models behave and the patterns for building on them. Anything tied to a specific provider stays in a clearly marked code block, so when it changes you update one line and the ideas around it still hold.

JunoA fast-moving field Model names and prices change almost monthly, so I don't memorise them and neither should you. What holds still is how these models behave and the patterns you build around them, and that's what we cover. Anything provider-specific lives in a marked code block, so a model release is a one-line change, not a rewrite.

JunoA fast-moving field Model names and prices churn almost monthly, which is why I stopped getting attached to any of them years ago. The behaviour and the patterns are what last, so that is where we spend our time. Anything tied to a provider sits in a clearly marked code block, so when the next release lands you change one line and move on.

Meet your guides

This handbook is hosted by three engineers from Crooked Orbit, and you pick which one shows you around. The material is the same for everyone; your guide sets the voice it is told in, and on the pages that adapt, how deep each section goes.

Switch guide any time from the picker at the top of any page, or open Preferences at the bottom right to change your guide, language, or font size.

When you're set, the handbook starts in earnest. The next chapter, A short history of AI, traces where these systems came from over seventy years, because that history is what tells you which parts of the field are durable and which are only this year's noise.