Learning path

Every lesson is one short video plus the exact code from the video. The best way to learn from them:

Watch the video once, without pausing.
Run the code and check that you get the same output as the video.
Read the lesson README. It walks through the code line by line, using the line numbers you saw on screen.
Break it. Do the “Try this” exercises at the bottom of each README. Changing a number and predicting what happens teaches more than reading.

Model Anatomy, in order

The lessons build on each other. If you are new, go in this order.

#	Lesson	You will understand
00	What is machine learning	What “learning from examples” means, in 8 lines.
01	Gradient descent	How every model improves: feel the slope, step downhill.
07	Linear regression, no libraries	The full training loop written out by hand, and what a bad learning rate does.
02	K-means	How a program finds groups in data that has no labels.

Lesson 07 comes right after 01 on purpose: it repeats lesson 00 with no NumPy, so you see every step.

#	Lesson	You will understand
08	Decision trees	How a model picks its questions, and why you can read its decisions.
09	A neural network from scratch	Layers, a non-linearity and backpropagation, on the XOR puzzle.
10	Overfitting	Why a perfect training score is a warning, and how validation catches it.
11	Choosing a model	How teams pick an algorithm: data first, goal second, model last.

#	Lesson	You will understand
03	The AI agent loop	Think, act, observe, repeat: how agents use tools.
06	RAG from scratch	How an assistant answers from your own documents.
04	Ontologies	How an agent connects facts across systems.
05	Evals and loop engineering	How to improve an AI system with measurements instead of guesses.

#	Lesson	You will understand
12	ETL vs ELT	Where the transform runs, and what each choice keeps.
13	PySpark	Partitions, lazy plans and shuffles: how Spark handles data too big for one machine.
14	Polars	How a lazy query reads only the columns and row groups it needs.

Build Lab episodes are projects. Each one builds a small, real piece of software across several parts.

Project	Parts
01 · API to Parquet	Part 1: fetch, clean, write Parquet, query with DuckDB
02 · Documents to data	Part 1: OCR invoice images into JSON, Parquet and SQL. Part 2: embeddings in LanceDB and retrieval for RAG

Python basics: variables, lists, loops, functions. If you can read a for loop, you can follow every lesson.
No math background needed. Where math shows up (a slope, an average, a distance), the README explains it in words first.

Words you don’t know are in the glossary. Want to know what’s coming next? See the topics.