Free preview
Learn AI online: free lessons on LLMs, RAG and AI agents
Learn AI the hands-on way. 33 modules take you from your first LLM call to RAG, AI agents and AI in production, with labs that run in your browser.
Whether you want to learn RAG, learn how to build AI agents, or just learn how to use an LLM in your own code, these lessons start where a programmer is and go as far as running AI features in production. Every lesson is free to read, and you run the commands in a real lab in your browser with nothing to install.
An AI roadmap for programmers: how to become an AI engineer
If you can program, start at Series A and work through the series in order. Each step has free lessons you read whole, labs you run in your browser and tasks marked by hidden tests.
- Series A: code with an AI assistant and work with coding agents through specs, context files, tests and careful review, then call a model from code, ask for structured output, use tools, build a chatbot and answer questions from your own documents with RAG.
- Series B: build AI agents, test them with evals, watch cost and reliability, secure them against prompt injection and ship a feature.
- Series C: harness engineering, several agents working together, evaluating and securing agents, retrieval systems and governing change.
- Series D, any time after Series A: see how a language model works, from tokens and attention to a tiny model you train, write better prompts, work with images and sound, run and adapt models yourself, write an MCP server, design AI products people can trust, and let a model turn questions into SQL that you check before trusting it.
- Series E, after Series B: evaluate in depth, protect personal data, operate, manage change and incidents, and keep agents safe.
AI glossary
Short, plain definitions of 48 AI terms, from tokens to MCP, and 10 comparisons such as RAG vs fine-tuning, each linked to the lesson that teaches it.
Open the AI glossarySeries A: AI for software engineers
For a student or a new graduate who can program and use Git. Code with an AI assistant and work with coding agents (clear specs, context files, tests as guardrails, reviewing what they change), then call a model, ask for structured output, use tools, build a chatbot and answer questions from documents.
12 lessons
AI-assisted software engineering
Learn AI-assisted software engineering hands on: write good prompts, review and test generated code, protect secrets and decide what to delegate.
See the module6 lessons
Agentic software engineering: working with coding agents
Learn to work with AI coding agents hands on: task specs, context files, tests as guardrails, reviewing changes, CI steps and budgets.
See the module6 lessons
AI for developers: the bridge
Learn AI for developers hands on: how a model works, HTTP and JSON, keeping keys secret, reading API docs, streaming and cost.
See the module6 lessons
Calling language models from code
Learn to call language models from code hands on: messages, parameters, streaming, retries and a command-line assistant.
See the module6 lessons
Structured output from models
Learn structured output from models hands on: ask for JSON, check it against a schema, repair bad answers and turn text into records.
See the module6 lessons
Tool calling
Learn tool calling hands on: describe a function to a model, run the call loop, handle errors and ask a person before risky calls.
See the module6 lessons
Building a chatbot
Learn building a chatbot hands on: keep a conversation, trim it, write a system prompt, stream answers into a web page and cite sources.
See the module6 lessons
Retrieval-augmented generation
Learn retrieval-augmented generation hands on: embeddings, chunking, vector search, hit rate, answering with sources and when not to use it.
See the module
Series B: AI in production
For experienced developers. Build agents, test them with evals, watch cost and reliability, secure them and ship an AI feature.
6 lessons
Building agents
Learn building AI agents hands on: plan and stop a run, give it memory, hand work between agents, checkpoint it and speak MCP.
See the module6 lessons
Evals and testing for model-backed features
Learn LLM evals and testing hands on: golden sets, scoring rules, regression runs, a model as judge and a gate that fails a bad change.
See the module6 lessons
Cost, tracing and reliability
Learn cost, tracing and reliability for LLM features hands on: spans, caching, budgets, latency limits, fallbacks and model routing.
See the module6 lessons
Securing LLM applications
Learn LLM security hands on: prompt injection, risk mapping, tool allow-lists, redacting personal data and audit logs for AI applications.
See the module6 lessons
Architecture and delivery of AI features
Learn AI feature architecture hands on: an LLM gateway, job queues, images and documents, prompt or RAG or fine-tuning, and prompt releases.
See the module6 lessons
Case study: shipping an AI feature
Learn shipping an AI feature hands on: write the brief, measure a first version, read evals, trace an incident, write a post-mortem and a decision log.
See the module
Series C: advanced agent systems
For engineers who have finished series B or have the same experience. Harness engineering, several agents together, evaluating and securing agents, retrieval systems and governing change.
6 lessons
Harness engineering
Learn agent harness engineering hands on: context budgets, tool design, permission modes, subagents, undo and a small harness.
See the module6 lessons
Multi-agent systems
Learn multi-agent systems hands on: when to split work, orchestrators and workers, shared state, checking claims and failure modes.
See the module6 lessons
Evaluating agents
Learn evaluating AI agents hands on: score their steps, check the state they leave, compare runs fairly, test dialogues and gate a change.
See the module7 lessons
Securing agents
Learn AI agent security hands on: threat models, data exfiltration, confused deputies, untrusted tools, privilege separation and red teaming.
See the module6 lessons
Retrieval systems
Learn retrieval systems hands on: hybrid search, reranking, query rewriting, chunk quality, access filters and agentic retrieval.
See the module7 lessons
Governing and changing AI systems
Learn AI governance hands on: rate risk, design real human oversight, use feedback, change a model safely and review a feature.
See the module
Series D: applied AI
For anyone who can program. How a language model works (tokens, attention and a tiny model of your own), prompting, images and sound, running models yourself, adapting a model, writing an MCP server, designing AI products people can trust, and having a model turn questions into SQL that you check.
6 lessons
How language models work: transformers from scratch
Learn how a language model works from scratch: tokens, embeddings, attention, a transformer block, next-token training and sampling.
See the module6 lessons
Prompting and context craft
Learn prompting hands on: write clear prompts, teach with examples, ask for working, use patterns, test a prompt and keep a prompt library.
See the module6 lessons
Multimodal and voice input
Learn multimodal AI hands on: how images, pages and sound become numbers, prepare inputs, check what a model saw, and judge cost and access.
See the module6 lessons
Running models yourself
Learn to run models yourself, hands on: model memory, quantising, serving and batching, licences, choosing a size and self-host or rent.
See the module6 lessons
Adapting a model
Learn model adaptation hands on: prompt first, training data, leaks, how an adapter works, judging a tuned model and drift.
See the module6 lessons
Writing an MCP server
Learn to write an MCP server hands on: messages, the handshake, validated tools, resources and prompts, scopes and testing.
See the module6 lessons
Designing AI products for people
Learn AI product design hands on: set expectations, show uncertainty, collect feedback, hand over to a person, access and cost.
See the module6 lessons
Text to SQL with checks
Learn text to SQL hands on: describe a schema for a language model, check its SQL by reading and running it, keep it read only and score it.
See the module
Series E: running and trusting AI
For engineers who have finished series B. Evaluate AI systems in depth, protect personal data, operate AI features, manage change and incidents, and keep agents safe.
6 lessons
Evaluating AI systems in depth
Learn to evaluate AI systems in depth: golden sets, regression tests, judge pitfalls, hallucination rate, retrieval metrics and synthetic eval data.
See the module6 lessons
Privacy and data protection for AI apps
Learn privacy for AI apps hands on: find and redact personal data, log safely, delete on request and review a feature's data flow.
See the module7 lessons
Operating AI features
Learn to operate AI features hands on: context budgets, caching and cost, limits and retries, graceful failure, latency and a runbook.
See the module6 lessons
Change and incident management for AI
Learn change and incident management for AI hands on: version pinning, canary rollout, A/B tests, incidents, audit trails and reviews.
See the module6 lessons
Agent safety in depth
Learn agent safety hands on: allowlists, a confined folder, snapshots and rollback, agent memory, checking steps and a design review.
See the module
How the lessons work
- Free to read: every lesson is open with no account.
- Hands-on: run the commands in a lab in your browser. Nothing to install.
- Checked: Python assignments with hidden tests mark your work.
- In your language: English, French, Spanish, German and Japanese, each written for that language.
Pay as you go, not a subscription
A free account gives you credits every month for labs and a few assistant questions a day. Want more AI help? Top up from $1. There is no subscription and no monthly fee, and top-up credits stay valid for at least a year.
Questions about learning AI here
Can I learn AI online for free?
Yes. Every lesson in the five series is free to read with no account. A free account lets you run the commands in a lab and ask the assistant a few questions a day.
Do I need machine learning or advanced maths?
No. The lessons are for people who can program. They teach how to use language models in software, so they need no machine learning maths. The one module on how a model works inside keeps the maths to multiplying and adding numbers in a table.
What is RAG and where do I learn it?
RAG, retrieval-augmented generation, lets a model answer from your own documents. Series A has a module that builds document Q&A, and Series C goes deeper into hybrid search, reranking and agentic retrieval.
Will I learn to build AI agents?
Yes. Series B starts with building agents, and Series C covers harness engineering, multi-agent systems, and evaluating and securing agents.
Is it a subscription?
No. It is pay as you go. Reading the lessons is free, and you only top up if you want more AI assistance. There is no monthly fee.
Which languages are the lessons in?
English, French, Spanish, German and Japanese. Each language is written on its own, not machine-translated from another.
Should I fine-tune a model or use RAG?
Try a better prompt first, then retrieval, and only then adapt a model. Series D has a module that shows how to judge whether training helps, and Series A and C teach the retrieval side.
Can I run an LLM on my own computer?
Series D has a module on running models yourself: how much memory a model needs, what quantising changes, serving and batching, licences, and when to host your own or rent. It teaches the sizing and cost decisions with worked numbers in the browser lab; it does not install a model on your machine.
What is MCP, the model context protocol?
MCP is a standard way for a model to reach tools and data. Series D has a module where you write an MCP server: the messages, the handshake, tools that check their input, resources and prompts, who may call what, and testing.
What is prompt engineering and can I learn it here?
It is writing the instructions and examples a model works from, and testing that a change really helps. Series D has a module on it: clear prompts, examples that teach, asking for working, patterns, testing a prompt and keeping a prompt library.
How do I become an AI engineer?
Learn to program, then follow the roadmap on this page: call models from code, build with retrieval and tools, test and secure what you build, and run it in production. The lessons teach the skills. They do not give a qualification.
Do I need a GPU, a paid API key or a powerful computer?
No. The labs run in your browser, and every command also works with no model by showing a clearly labelled recorded sample. A free account's credits cover the lab, and you can top up if you want more.
How do I test and evaluate an LLM application?
Series B has a module on evals: fixed cases, a rule for each and a before and an after. Series C goes on to evaluating agents, and Series D shows how to test a single prompt.
How do I protect an AI app from prompt injection?
Series B has a module on securing LLM applications, including prompt injection, and Series C covers securing agents. The lessons never attack anybody's system.
Teaching a class? Set this module for your students with a teacher guide, marked assignments and a lab for every student.