Edition · November 2026

AI Engineer Roadmap

160 concepts across 8 areas, the order to learn them in, and one thing to build after each area. Tick a concept only after you have used it in code.

8 areas20 concepts each~26 weeks10–12 h / week
0 / 160 used in code
to learnused it
01
02
03
04
05
06
07
08
Hover a square to see the concept. Click it to mark it as used. Click an area name to open it below.
Saved in this browser only.
Start here

How to use this roadmap

Each area has 20 concepts, each with a full lesson. Learn a concept, then use it in code the same day. Tick the box only when you have used it, not when you have watched a video about it.

Before you start

  • Python. Functions, classes, virtual environments, pip, reading other people's code.
  • Git and GitHub. Commit, branch, pull request. Every project in this roadmap goes on GitHub.
  • Basic SQL. SELECT, JOIN, INDEX. You will store everything in Postgres.
  • One LLM API call. Get an API key from any provider and print a reply from Python. That's enough to begin.

Three rules

  1. 01–10 first. In every area the first ten concepts are the base. Come back for 11–20 after you have shipped something.
  2. Build after every area. Each area ends with one project. Six small shipped projects beat sixty tutorials.
  3. Measure from Phase 4 on. Once you know evals, every prompt or model change gets tested before it goes live.
Suggested order · about 6 months at 10–12 hours a week

The 26-week route

Each stop is drawn to its length in weeks. Pick a stop to open its area.

The 8 areas

Learn it, use it, tick it

Area 01 / 08 · LLM

LLM

Phase 1 · Weeks 3–6

How the model works inside, and how to talk to it so you get what you want.

01–10 · The base

Learn these first.

  • 01TokenizationText is cut into tokens (word pieces) before the model reads it. Limits and prices are counted in tokens.
  • 02Transformer ArchitectureThe network design behind GPT, Claude and Llama: stacked attention and feed-forward layers.
  • 03Self-AttentionEvery token looks at the other tokens and decides which ones matter for its meaning.
  • 04Context WindowsThe most tokens (prompt + answer) a model can handle in one call.
  • 05KV CacheSaved keys and values of earlier tokens, so each new token doesn’t redo the whole prompt.
  • 06Temperature and Top-pSampling knobs. Low values give predictable answers, high values give varied ones.
  • 07System PromptsStanding instructions that set the model’s role, rules and output style.
  • 08Few-shot PromptingPut 2–5 worked examples in the prompt so the model copies the pattern.
  • 09Chain of ThoughtAsk the model to reason step by step before it answers. Helps on multi-step problems.
  • 10Reasoning ModelsModels trained to think at length before answering. Slower and pricier, better on hard tasks.

11–20 · Production level

Come back after you have shipped something.

  • 11Structured OutputsForce the reply into JSON that matches your schema, so your code can parse it every time.
  • 12Prompt CachingReuse an unchanged prompt prefix across calls to cut cost and latency.
  • 13Token PricingInput, output and cached tokens cost different amounts. Know your cost per call.
  • 14HallucinationsConfident but wrong answers. Reduce them with grounding, citations and permission to say ‘I don’t know’.
  • 15Multimodal InputsSending images, PDFs or audio to the model along with text.
  • 16Fine-tuning with LoRATrain small adapter matrices instead of all the weights. A cheap way to specialise a model.
  • 17QLoRALoRA on top of a 4-bit quantized base model, so fine-tuning fits on one GPU.
  • 18QuantizationStore weights in fewer bits (8-bit, 4-bit) to save memory and run faster, at a small quality cost.
  • 19DistillationTrain a small ‘student’ model to copy the answers of a big ‘teacher’ model.
  • 20Open vs Closed ModelsOpen weights you can host yourself (Llama, Qwen, Mistral) vs API-only models (GPT, Claude, Gemini).
BUILD THIS

A command-line chatbot with a system prompt, streaming replies, a JSON output mode and a live token + cost counter for every message.

Project guide →
Finish line

Your portfolio at the end

After the 8 areas, turn your small builds into one or two complete projects. Recruiters and clients look for a live link, a README that explains your choices, and numbers from your evals.

Docs assistant for a real company

RAG over 200+ real documents, hybrid search, citations, an eval suite with 50 questions, deployed behind auth.

RAGEvalsDeploymentSecurity

Agent that does a real job

An agent with 3–5 tools (one of them your own MCP server), approvals for risky steps, traces and a step budget.

AI AgentsEvalsSecurity

AI gateway

One API in front of several models: routing, fallbacks, semantic cache, per-user budgets and a cost dashboard.

System DesignBackendDeployment
Live demoA live link and a short video of it working.
Eval resultsWhat you measured, and the score before and after your changes.
Cost and latencyPer request, and how you brought them down.
SecurityInjection tests, permissions, secrets.