Level 03 · Technical

Building With AI

You write code or work with data. This track moves past "how to use AI tools" into how the systems underneath them actually work.

Last updated: September 2026

What this track covers

How large language models are structured — tokens, embeddings, and the transformer architecture, in plain terms.
Calling AI models via API — the difference between a chat interface and building your own application on top of a model.
Prompt engineering as a discipline — few-shot examples, system prompts, and structured output.
Working with embeddings and retrieval — how tools "search" documents and ground answers in real data (RAG).

A useful mental model: a language model is a very sophisticated autocomplete — it predicts the next most likely token given everything before it. Almost everything else in this track builds on that one fact.

Where to go from here

If you're already shipping AI features in production, or making build-vs-buy and cost decisions at an organisational level, the Pro/Advanced track covers that ground.

Go deeper with these lessons

Build understanding before building features

The technical track is deliberately not a “copy this code and ship an AI app” course. APIs change quickly. The concepts underneath them last longer. I would first make sure tokens, context, inference, embeddings and retrieval make sense, then use the documentation for whichever model provider you actually choose.

A useful project sequence

Call a model. Build the smallest possible API request and inspect the raw response rather than hiding it behind a framework.
Control the output. Experiment with system instructions, structured output and a few examples. Record what fails as well as what works.
Ground an answer. Add retrieval over a small set of documents and check whether the generated answer is supported by the retrieved text.
Evaluate. Create a fixed set of test questions before changing prompts or models. Otherwise it is easy to mistake a different answer for a better one.

Technical checkpoint

You are ready for the Advanced track when you can explain the data flow through your application, identify where hallucinations or privacy problems can enter, and test a model change against repeatable examples instead of intuition alone.