Machine Learning from Zero
by Sergey Okinchuk
Ch. 1

Machine Learning from Zero

Welcome

Machine learning is everywhere you already are — the spam filter, the recommendations, the chatbot — yet it is usually explained either as magic or as math. This course does neither. It builds a working mental model of ML from the ground up, without a single line of code: what a model really is, how training works, why models fail, how to measure them honestly, how a model becomes part of a product, and how the same ideas scale up into the generative AI you talk to. By the end, you can follow a data-team conversation, judge whether an idea is an ML problem at all, and know exactly what to learn next.

11 chapters 53 topics covered 7 hours audio Knowledge check on every topic

About This Course

"Machine learning" has become the most name-dropped and least understood phrase in tech. One camp presents it as near-sorcery; another buries it under calculus by the second page. Meanwhile the actual idea — a computer working out its own rules from examples, because nobody can write the rules by hand — is something anyone can genuinely understand, limits and all. That understanding, complete and honest, is what this course delivers.

To keep it concrete, the whole book follows one company and one person. Plateful is a food-delivery app with the classic prediction problems: when will the order arrive, which reviews are fake, who is about to stop ordering, is this payment fraud. Iris is the product manager who joins Plateful knowing nothing about ML and learns, chapter by chapter, to work with its data team. She is the reader's stand-in: she never writes code, and she never needs to — by the end she is the person in the room asking the questions that actually matter.

By the end, the words that fill every AI headline and every data-team conversation — model, training, overfitting, accuracy, precision, deployment, neural network, LLM, hallucination — will mean something specific to you, and you will know how they fit together. There is no code here and no math beyond arithmetic. This course teaches you to understand machine learning; learning to build it is a different road, and the final chapter maps it honestly.

Who This Is For

Anyone who wants to genuinely understand machine learning rather than nod along to it: career-switchers deciding whether this field is for them, students who want the ideas before the formulas, managers and product people who work next to a data team, and anyone who talks to a chatbot daily and wonders what is actually happening in there. It assumes no engineering background at all — only everyday computer literacy. If you already train models for a living, this course is below you; if you have ever wondered whether it is too late or too hard to understand this field, it was built for exactly you.

What You Should Already Know

  • How to use a computer, a browser, and apps as an everyday user — nothing more technical than that
  • No programming, no command line, and no mathematics beyond everyday arithmetic
  • No prior exposure to AI or ML — every term is defined the first time it appears
  • A willingness to trade magic for understanding; the trade is worth it

How the Course Is Built

The eleven chapters build in a deliberate order. First the core idea — learned rules versus written rules, and what a model physically is. Then the raw material: data, where it comes from, and why its quality sets the ceiling on everything. Then how learning actually works — training, overfitting, and the honest exam every model must sit. Two chapters sort the kinds of learning (with answers, and without), and one teaches the skill this book prizes most: measuring models honestly. Then the full journey from business idea to deployed, aging, maintained product. Only then do we scale up — neural networks, and the generative AI everyone is talking to — before facing the real world: fairness, privacy, explainability, and what ML means for your job. The final chapter is a map of everywhere you can go from here.

Every topic has the same gentle shape: a hook from Iris's world to start, the idea built step by step, one everyday comparison to make it stick, the mix-ups people usually run into, why it matters, and a short knowledge check. It is patient, but it keeps moving — you are here to learn, not to be slowed down.

Understand, don't memorize
No formulas to recite and no tool menus to learn. We build the why behind every idea — what problem it solves, where it breaks — so you can reason about systems you have never seen before.
One company, all the way through
Plateful's problems accumulate chapter by chapter — the delivery predictor, the fraud filter, the churn model, the chatbot — so every new idea lands on a story you already know.
Honest, never hyped
Neither "AI understands you" nor "it's all just statistics". What these systems genuinely do, how well, and where they fail — stated plainly, including for the shiniest ones. Nothing you will have to unlearn.
Concepts that travel
Products and model versions churn; mechanisms don't. This course teaches the mechanisms — prediction, training, evaluation, generation — that will still explain the AI news five years from now.

Chapter Map

Chapter 1
What Machine Learning Is
The core idea: rules learned from examples instead of written by hand. The AI/ML/deep-learning/GenAI map, the ML already around you, what a model physically is, and when not to use ML at all.
Chapter 2
Data: The Raw Material
Rows, features, and labels; the kinds of data and why everything becomes numbers; where data and labels actually come from; why cleaning is the real work; and garbage in, garbage out.
Chapter 3
How a Model Learns
Training as guess-measure-adjust; what the model looks at; the train/test split; overfitting — the central disease of ML — and the first honest look at "how good is good enough".
Chapter 4
Supervised Learning
Learning from examples that carry answers: classification and regression, a tour of simple models that still run the world, and why committees of models beat soloists.
Chapter 5
Learning Without Answers
What can be learned when no answer column exists: clustering that finds customer segments by itself, anomaly detection that spots the odd one out, and learning by trial and reward.
Chapter 6
Measuring Models Honestly
Why 99% accuracy can be worthless, precision and recall, the confusion matrix and the price of each kind of error, measuring predictions of numbers, and the baselines that keep teams honest.
Chapter 7
The ML Workflow End to End
From a business wish to a precise problem, from raw data to telling features, training and tuning as a fair tournament, deployment into the product — and why models age and need care forever.
Chapter 8
Neural Networks & Deep Learning
Neurons as little weighers, why depth plus data plus GPUs changed everything, how machines learned to see and to read — and when deep learning is the wrong tool.
Chapter 9
Generative AI
Generation as prediction run forward: how LLMs are made in three stages, why they hallucinate and how to work with them honestly, image generation, and the honest map of what they can and cannot do.
Chapter 10
ML in the Real World
Where models meet people: how bias gets in without a villain, privacy on both sides of the prompt, the right to an explanation, the limits of prediction, and what ML actually does to jobs.
Chapter 11
Where to Go Next
The ML job map, how models run in production (and which courses teach that world), the honest syllabus for learning to build, and the mental model you carry out of this book.

Disclaimer

This course is an independent educational project created and maintained by Sergey Okinchuk. It is provided for learning and reference purposes only.

No affiliation. This course is not affiliated with, sponsored by, endorsed by, or officially connected to any company, product, or organization mentioned. All opinions, interpretations, and recommendations expressed are those of the author. Plateful and all characters in this course are fictional; any resemblance to real companies or persons is coincidental.

Trademarks. Product and company names referenced are the property of their respective owners. Use of these names is for identification and educational purposes only and does not imply any endorsement.

Educational simplifications. This material teaches durable concepts for understanding, not operational instructions. Explanations are deliberately simplified for learning; where a simplification could mislead, the text flags it. This course is not professional, legal, medical, or financial advice, and is not a substitute for qualified expertise in any decision involving machine-learning systems.

Accuracy and currency. The field evolves quickly — capabilities, products, and practices drift over time. Facts in this course reflect the author's understanding at the time of writing and may not be current. The course deliberately teaches version-free mechanisms rather than product specifics; always consult authoritative sources for the current state of any tool or model.

No warranty. This material is provided "as is" without warranty of any kind. The author accepts no liability for any loss or damage arising from reliance on the content.