How LLMs Work
Learn what happens between sending a prompt and getting a reply, and how to build with it.
What you'll learn
- How models split text into tokens and predict words
- Why models make things up, and how to ground them
- How temperature, top-p and reasoning shape replies
- How to estimate the cost and latency of an AI feature
- How to pick the right model by testing it on your own data
- When a feature needs a model, and when plain code is better
Needs basic programming experience in any language
Course content
About this course
Most developers now use large language models every day, but few know what happens between sending a prompt and getting a reply. This course explains how LLMs actually work, in plain language and without maths, and ties every idea back to decisions you make when you build with them.
You will see how text becomes tokens and why that drives cost and context limits, how a model predicts one token at a time and why it can't take a word back, how training shapes what models know and why they make things up, and what sampling settings and reasoning models actually change. You will then learn how to estimate cost and latency, make features feel fast, and pick the right model by testing on your own data.
Course by
Kamran Ahmed
Kamran is the founder of roadmap.sh. He has spent over a decade building products at startups and scale-ups, and created open-source tools used by millions of developers. He loves teaching and simplifying complex topics.
Published 2d ago
