\( \newcommand{\vect}[1]{\mathbf{#1}} \newcommand{\mat}[1]{\mathbf{#1}} \newcommand{\R}{\mathbb{R}} \newcommand{\E}{\mathbb{E}} \newcommand{\Prob}{\mathbb{P}} \newcommand{\T}{^{\top}} \newcommand{\norm}[1]{\lVert #1 \rVert} \newcommand{\abs}[1]{\lvert #1 \rvert} \newcommand{\ihat}{\mkern1mu\hat{\boldsymbol{\imath}}\mkern2mu} \newcommand{\jhat}{\mkern1mu\hat{\boldsymbol{\jmath}}\mkern2mu} \newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\argmax}{\operatorname*{arg\,max}} \newcommand{\argmin}{\operatorname*{arg\,min}} \newcommand{\cdata}[1]{\class{c-data}{#1}} \newcommand{\cmodel}[1]{\class{c-model}{#1}} \newcommand{\ctruth}[1]{\class{c-truth}{#1}} \newcommand{\cgrad}[1]{\class{c-grad}{#1}} \newcommand{\closs}[1]{\class{c-loss}{#1}} \newcommand{\cparam}[1]{\class{c-param}{#1}} \newcommand{\csecond}[1]{\class{c-second}{#1}} \renewcommand{\,}{\mkern3mu} \)

About

Why Applied AI Mastery exists, the eight design principles behind every course and the research they rest on, how lessons are verified, and how the material is licensed.

Applied AI Mastery is an open repository of courses in data science and AI. It aims to take a learner from school-level maths to research-level AI on one connected path, and to explain every idea with a picture before an equation.

It is created and maintained by Michael Kusi-Appiah, in the open, on GitHub.

What makes it different

• One connected path. Today a learner stitches maths, machine learning, deep learning and reinforcement learning together from separate books and sites. Here every lesson declares its prerequisites, and one notation runs through everything.
• Pictures carry the teaching. Each concept gets a diagram before it gets an equation. The best visual sites do this for a few dozen topics. The aim here is to do it for every topic.
• Depth without gaps. Every derivation is written out in full, and every algorithm is built from scratch in code before a library is used.

The eight design principles

Each rule rests on published evidence or on a gap in what already exists.

1. Five levels with exit skills. Every course carries a level from L0 to L4, so you know what you must already know and what you will be able to do at the end.
2. A prerequisite graph, not a reading list. Every lesson names the lessons it depends on.
3. Picture before equation. In Mayer’s review of multimedia learning, people who learned from words and graphics beat those given words alone across 13 experimental comparisons, with a median effect size of 1.35.
4. Static first. Motion only when motion is the idea. A meta-analysis of 181 studies found that text plus diagrams gave large, consistent gains, while animation, games and simulations were less consistent.
5. No decoration. Mayer reports median effect sizes of 0.86 for removing extraneous material, 0.70 for highlighting essentials, and 0.82 for placing labels next to the parts they describe. Every mark on a figure must teach.
6. Retrieval and spacing built in. A review of ten study techniques rated practice testing and distributed practice highest. Every lesson ends with recall questions, and questions return in later lessons.
7. From scratch, then the library. Learners build each algorithm in plain NumPy first. Only then do they use scikit-learn or PyTorch.
8. Verified before published. Every claim has a source, every derivation has a second check, and every code cell runs in automated tests.

How a lesson is checked

Fluent text can hide wrong maths and invented citations, so nothing here is trusted because it reads well.

• Machines check what machines can. On every change, automated tests recompute every stated number, run every code cell, regenerate every figure, check every internal link, and check each lesson against the blueprint and the notation guide.
• People check the rest. A lesson is marked in review until a named person has signed off its sources, maths, code, figures and clarity. Drafting tools may help write a lesson. They cannot approve one.
• Sources are opened. Every citation is opened and read against the claim it supports. A citation that cannot be opened is removed.
• Corrections are public. Mistakes are fixed through the issue tracker and recorded in the changelog.

The full rules are in the style guide.

The illustrations

Every figure on this site is drawn by a script in the repository. There are no screenshots and no borrowed images. One colour language runs through all of them: blue always means data, vermillion always means a model’s output, green always means a gradient or an update. The colours are the Okabe-Ito palette, chosen because it stays distinct under the common forms of colour-vision deficiency, and the typeface is Atkinson Hyperlegible, designed by the Braille Institute for legibility.

Free, and private

• Every lesson is free to read and will stay free.
• There are no accounts, no cookies and no analytics on this site. Lesson progress is stored in your own browser and never leaves it.
• On pages with maths, your browser loads the MathJax typesetting library from jsDelivr and a small compatibility script from cdnjs. Those are the only requests this site makes to third parties.

Licence

• Lessons, text and figures: Creative Commons BY-NC-SA 4.0. You may copy, share, translate and adapt them for any non-commercial purpose, as long as you give credit and share your version under the same licence.
• Code (figure scripts, tools, and code shown in lessons): MIT.

Teachers: you are welcome to use the lessons and figures in your classes. To credit a figure, keep its address (for example M02.4.1-F1) and link back to the lesson.

How to cite

Michael Kusi-Appiah. Applied AI Mastery. 2026. https://applied-ai-mastery.com

Get in touch

Questions, corrections and ideas are welcome in GitHub Discussions and the issue tracker.