From the Perceptron to LLMs

A rigorous, math-first course on deep learning, from linear models up to modern large language models. Sessions are short, dense in derivations, and meant to be followed with a pen in hand.

The intended reader is a STEM graduate (physics, EE, applied math, CS) with multivariable calculus and linear algebra in the bank. Every algebraic step is shown — no “it follows that”, no “it is easy to see”.

How to start

Use the sidebar on the left to navigate the chapters. Mark sessions as complete and a ✓ appears in the index.

Start with the first session →

About

Hi, I'm Luan — a data scientist with an MSc in physics. I built this website with Claude Opus 4.6. During my master's I studied the mathematical theory of neural networks and image recognition with CNNs (see my dissertation). Years later, wanting to understand deeply how LLMs work, I decided to write this course on top of that foundation.

If you'd like to contribute or have any feedback, reach me on LinkedIn or by email at luanmisaelmoura@gmail.com.

MVP demo. Progress is stored in your browser'slocalStorage — clearing site data resets the checkmarks. Cross-device sync via Google login is planned for a later phase (see website_plan.md).