From writing code to shipping AI systems.
6 stages, 26 topics, about 15 hours end to end. Each topic comes after the ones it depends on, so you never meet an idea before the one that explains it. Reading is free.
Pick the line that sounds like you.
Start at that stage. Skip the ones above it; you can always go back.
- 1
Work with AI today
Use a coding agent on a real repository and trust what it hands back.
Start here if: You write code and have not built anything on an LLM yet.3 topics, about 1 h
- 1.1Working With Coding AgentsThe loop that gets trustworthy code from an agent: explore, plan, implement, verify and review, with specs, small diffs and parallel work done safely.30 minfoundational
- 1.2AGENTS.md & Repo InstructionsThe small file that stops you repeating yourself: what belongs in AGENTS.md, what to cut, how to share it across tools, and when a rule needs a real check.25 minfoundational
- 1.3Reviewing Agent-Written CodeAgent mistakes are fluent, so review changes too: read tests first, check every dependency, separate generation from evaluation, and scale depth by risk.30 minintermediate
- 2
Know how models behave
Call a model from code, get structured output back, and say what the request cost.
Start here if: You use AI tools daily and want to call a model from your own code.5 topics, about 2 h
- 2.1Context WindowsUnderstand context length limits, lost-in-the-middle effects, and strategies for working within or around token budgets.20 minfoundational
- 2.2System PromptsCraft the foundational instruction layer that shapes model behaviour, tone, and capability boundaries for every conversation.25 minfoundational
- 2.3Token EconomicsUnderstand how token pricing, context windows, and prompt engineering choices translate directly into your monthly AI infrastructure bill.35 minintermediate
- 2.4JSON ModeForce structured JSON output from LLMs — using native JSON mode, function calling schemas, or prompt-based extraction.30 minintermediate
- 2.5Function CallingDefine typed function signatures that models can invoke — the bridge between natural language and programmatic action.35 minintermediate
- 3
Ground it in your data
Build a RAG pipeline over your own documents and measure whether retrieval is the part that fails.
Start here if: You have shipped a prompt and now need answers from your own data.4 topics, about 3 h
- 3.1EmbeddingsConvert text, images, and code into dense vectors that capture semantic meaning — the foundational primitive of modern AI systems.35 minintermediate
- 3.2Vector SearchUse approximate nearest-neighbour algorithms to search semantic space — the retrieval engine that makes RAG actually work.40 minintermediate
- 3.3Retrieval-Augmented GenerationGround LLM outputs in your own data by retrieving relevant context at query time and feeding it to the model.45 minintermediate
- 3.4LLM EvaluationMove beyond vibe-checks — build systematic offline and online eval pipelines that give you confidence before and after every deployment.40 minintermediate
- 4
Let it act
Design an agent that calls tools, plans across steps and keeps the context it needs.
Start here if: You have a working RAG system and want the model to take actions.5 topics, about 4 h
- 4.1Agent ArchitectureDesign the internal structure of an AI agent — its memory, tools, planner, and executor — to handle complex multi-step tasks reliably.50 minadvanced
- 4.2Tool UseEquip LLMs with callable functions — code executors, web search, APIs — and design schemas that make tool selection reliable.40 minintermediate
- 4.3MCP IntegrationConnect agents to external systems via Model Context Protocol — discover, register, and invoke MCP tools from inside agent loops.40 minadvanced
- 4.4Planning LoopsBuild reliable thought-action-observation cycles that let agents decompose complex goals into executable steps without going off-track.45 minadvanced
- 4.5Short-Term MemoryManage the context window as working memory — what to include, summarise, or drop to keep agents coherent across long conversations.35 minintermediate
- 5
Prove it works
Catch a quality regression before users do, and trace any bad answer back to the step that caused it.
Start here if: You have an agent or a RAG system in front of real users.5 topics, about 3 h
- 5.1Offline EvaluationsPre-deployment evaluation — golden datasets, automated scoring, regression testing for LLM applications.30 minintermediate
- 5.2Online EvaluationsProduction-time evaluation — user feedback, A/B testing, and real-time quality monitoring for LLM outputs.25 minintermediate
- 5.3OpenTelemetry for AIInstrument LLM apps with OpenTelemetry — semantic conventions for gen-ai, distributed tracing, and vendor-neutral observability.30 minintermediate
- 5.4Agent ObservabilitySee why an agent did what it did: one trace per run, OpenTelemetry GenAI spans, cost per task, and evals that catch what green dashboards miss.35 minintermediate
- 5.5Guardrails & SafetyLayer input/output validation around LLMs to prevent harmful outputs, prompt injection, and compliance violations in production.40 minintermediate
- 6
Ship it and keep it running
Pick models by cost and latency, serve them, defend the prompt, and survive a failed run.
Start here if: You own the system in production and the bill that comes with it.4 topics, about 3 h
- 6.1Cost-PerformanceCompare $/M tokens, latency, and quality across models — choose the best model for your budget and requirements.25 minintermediate
- 6.2Model ServingServe LLMs at production scale — batching, quantisation, speculative decoding, and GPU memory management for low-latency inference.50 minadvanced
- 6.3Prompt InjectionUnderstand and defend against adversarial inputs that hijack model instructions — the #1 LLM security vulnerability.35 minintermediate
- 6.4Durable Agents & SandboxingAgents that survive crashes without repeating themselves: checkpoints, idempotent side effects, approvals that hold no worker, and sandboxes for generated code.40 minadvanced
Prove it with a certification
Which AI certification fits your role, with a mock exam for each and every answer explained.
Find your certificationNeed one answer, not a course?
Every topic on the site by domain, including system design, databases, APIs, cloud and testing.
Browse all topicsCheck what changed
What to adopt, trial, assess or hold across AI techniques, platforms and tools.
Open the radar