Support agents lose time searching 2,000 help-centre pages. You build an assistant that drafts a cited answer from the docs, hands off to a human when unsure, and stays within a cost budget.
An AI engineer builds products on top of foundation models: chat, search, agents, document workflows. You rarely train models. You design the system around them: prompts, retrieval, tools, evals, guardrails, cost and latency.
Goal: Go from no programming background to building and shipping a small, working app that calls a language model, and explain in plain words what the model does and where it fails.
Before you start
Beginner · 8 weeks
Week 1: What an AI engineer is and how a computer runs code
What an AI engineer builds versus what a researcher builds · Vocabulary: model, prompt, token, API, inference, training · Terminal basics: cd, ls, mkdir, running a script · Install Python and a code editor
Practice: Write and run hello.py that prints your name and today's date. Write a one-page glossary in your own words.
Practice: Write and run hello.py that prints your name and today's date. Write a one-page glossary in your own words.
Common pitfalls
Think first, then reveal
What is a token?
Think first, then reveal
Why do you keep an API key in an environment variable and not in the code?
Think first, then reveal
Why might a model give a confident wrong answer?
Think first, then reveal
What does temperature do?
Think first, then reveal
Your script crashes with a traceback. What do you do first?
Help pages are cleaned, split into chunks and tagged with product and version.
Turning messy source documents into clean, chunked text with metadata.
Why this pick
Retrieval quality is capped by chunk quality; most wrong answers start here.
What the role must know
Beginner mistake
Core
Python
LLM systems
Quality and safety
Think first, then reveal
What is a token?
Think first, then reveal
Why do you keep an API key in an environment variable and not in the code?
Think first, then reveal
Why might a model give a confident wrong answer?
Think first, then reveal
What does temperature do?
Think first, then reveal
Your script crashes with a traceback. What do you do first?
Think first, then reveal
How do you choose chunk size for retrieval?
Think first, then reveal
Your RAG app gives a wrong answer. How do you find the cause?
Think first, then reveal
How do you stop a model returning invalid JSON?
Think first, then reveal
What is prompt injection and how do you reduce it?
Think first, then reveal
How do you know a prompt change is an improvement?
Think first, then reveal
When would you not build an agent?
Think first, then reveal
How do you trust a model-graded eval?
Think first, then reveal
Design the failure handling when your model provider has an outage.
Think first, then reveal
How do you secure an agent that can read email and call tools?
Think first, then reveal
How do you control cost without hurting quality?
Build one, measure it and put it on your CV. Each has tools, setup commands and milestones.