AI/ML from the Ground Up

Hi — I'm Sean Egan. A machine-learning obsessive who got tired of explanations that stop at the hand-waving. This is the second of my three explainer sites. The first, How LLMs Actually Work, takes one famous kind of model apart down to the bolts. This one answers the bigger question: what is this whole field, really? The third, How AI Makes Images, takes one branch of that field — machines that generate pictures — and follows it all the way to the bottom.

This site stands alone — you don't need the LLM site first, and every piece of math is rebuilt here from the ground up. The goal is bolder than the first site's, though. By the end, you won't just understand how these methods work; you'll be able to look at a fresh problem — mapping a network of contacts, spotting fraud, sorting photos — and say, with reasons, "this calls for a graph neural network" or "honestly, a random forest will beat anything fancy here." That judgment is what I mean by mastering the method.

Every technical field I have worked in taught me the same lesson: you don't understand a thing until you can predict what it will do. Same rule here. So: pictures first, then math, then your own predictions — checked against interactive demos on every floor of the building.

How this site works

Four parts. Part I makes "machine learning" mean something precise — a machine adjusting a function until it fits data — and builds the math toolkit. Part II constructs neural networks from a single artificial neuron and teaches them to learn. Part III is the field guide: the classical toolbox, and the architectures shaped for images, sequences, and graphs. Part IV is the capstone — you, choosing the right tool for six real problems, and defending the choice.

Every chapter follows the same rhythm: a picture, then the math (with every equation translated into plain English), then something to play with. Nothing is used before it is taught.

The three sites make a ladder, and you are on the second rung. Each one stands alone, but read in order they climb: this site maps the field and gives you the judgment to pick a tool; How LLMs Actually Work takes the slow, patient road through the mathematics and builds one language model from the first vector up — it is where several chapters here send you for the unhurried version of a proof; and How AI Makes Images is the newest, and the last: it takes the diffusion models you will meet in Chapter 11 and builds a working image generator out of them, one arithmetic step at a time. Finish here and either of the other two is a reasonable next climb.

Part I