Roadmap · 5 stages · 14 topics

AI Engineer

Build reliable features on top of large language models: context, retrieval, agents, evals, and shipping them safely.

Assumes you can already build a basic web service (the first stages of the Backend Engineer track). No machine-learning background needed: this track is about building with models, not training them.

0%complete

0 of 14 topics done

~10 h 30 min of reading left

Start: Text & Encodings →

Follow it and the end of each topic suggests the next one on this track.

Stage 10/3 done

Foundations

Models read and write text as numbers. Knowing how text is encoded, how HTTP calls fail and how to test code makes every later topic easier.

Code here comes in Go and JavaScript. New to Go? Start with the 90-second Go primer.

Optional side reads (not counted):

Stage 20/2 done

How models behave

An LLM predicts the next token. That one idea explains context limits, randomness and hallucinations, and how to write prompts that get structured, checkable output.

Stage 30/4 done

Giving models the right context

A model only knows what's in its training data and its prompt. Embeddings, vector search and RAG let it answer from your own documents, and fail in ways you need to recognize.

Optional side reads (not counted):

Stage 40/1 done

Agents and tools

When a model can call tools, it can act, not just answer. Learn the agent loop, MCP, and when a plain function call is the better design.

Optional side reads (not counted):

Stage 50/4 done

Ship it safely

Non-deterministic output needs its own kind of testing. Production adds streaming, caching, retries and cost, and anything that reads untrusted text can be attacked through it.

Optional side reads (not counted):