From a fitted line to a transformer, then training and serving it.
Not startedUnmark
Fit a model to data, then test it without fooling yourself.
How a network learns, one gradient at a time.
How text turns into vectors a model can use.
Attention, and the block every language model is built from.
How weights get made, and what fine-tuning changes.
Why answers are slow or costly, and how to fix it.
Prompts, evals, search and agents, then running it for real users.
Prompts and structured output you can rely on.
Before you ship a change, you need a number that says it helped.
How to find the right pages to show the model.
Give a model tools and a loop, then keep it inside safe limits.
Routing, budgets and fallbacks keep it all up once real people use it.