Learn how modern AI
actually works.

The mathematics and neural-network mechanics everything else assumes.

Open the box. How a language model actually turns text into predictions.

From raw web crawl to an aligned model — data, dynamics, scale and adaptation.

Classical RL, then the specific dialect of it that post-trains language models.

Where the model meets the silicon, the memory bus and the latency budget.

Building things people use: retrieval, prompts, agents and production architecture.

Models that think longer, the evals that measure them, and the failure modes that matter.

Beyond text — vision, speech, robotics and scientific discovery.

The statistics and non-neural models that still decide most production predictions.

Telling correlation from cause, and measuring whether a change actually helped.

Data with an arrow of time, where shuffling the rows destroys the problem.

Learning over relations, catalogues and columns rather than free text.

Diffusion, flows, adversarial games and state space models, and the theory that connects them.

Making a model smaller, cheaper and local without giving away the thing that made it useful.

Thirty years of ranking research that RAG rediscovered, usually the hard way.

The pipelines, formats and contracts that decide whether a model ever sees correct inputs.

Everything between a notebook that works and a system that keeps working.

Attacks on models, data and the supply chain, and the defences that survive contact.

Frameworks, regulation and audit evidence, treated as engineering rather than paperwork.

The people using the system, the product decisions around it, and what compute actually costs.