Introduction

This guide teaches how AI applications are built and how to explain the decisions behind them in an AI engineering interview. Later sections build on earlier ones, so read them in order when you can.

What is AI engineering?

AI engineering is the process of building applications on top of AI models that have already been trained. You don't train the model yourself. Rather, you adapt the model to your application, making sure that it's able to do its task correctly, using your own data and staying within the limits you set.

Consider an assistant answering a question about a document. You need to decide which information it can use, how that information reaches the model, and what should happen when the document does not contain an answer.

The interview

AI engineering interviews vary between companies, but most include some of these rounds:

• System design: design a feature such as a document assistant. Follow a question from the user to the model's answer, choose between approaches, and explain what each one gives up.
• Coding: build a small feature that calls a model's API, and handle what happens when the call fails or the output is wrong.
• Concepts: explain the ideas in this guide and say when each one helps.
• Past work: talk through an AI feature you built: the choices you made, what went wrong and what you'd change.

What this guide covers

1. 01Models & APIssend a request to a model's API and read its response, including streaming (receiving the answer as it's written) and structured output (getting JSON in a format you define).
2. 02Evalstests that check the model's output against criteria you define, such as whether every claim cites a source.
3. 03Prompt Engineeringwrite the instructions the model follows, choose what text to send with the question, and check the result with evals.
4. 04Tool Usedefine functions the model can ask your application to run, such as a search or a database lookup, and limit what they can do.
5. 05Open nowRAG →retrieval-augmented generationsearch your own documents for the passages that answer a question, and give those passages to the model to answer from.
6. 06Model Capabilitiessend images and documents, get answers that cite their sources, reuse repeated parts of a request to save cost and time (prompt caching), and let the model run code.
7. 07MCPModel Context Protocola standard way for an application to offer tools and data to a model.
8. 08Workflowschain model calls into fixed steps you can test one at a time.
9. 09Agentslet a model choose its own steps, only when a fixed workflow can't do the job.

Where to start

Only RAG is available so far, so start there. The other sections will be added in this order. Skip any section you already know.

After each section, explain the idea out loud as you would to an interviewer.