GuideIntermediate

AI System Design Interview Guide

Almost every AI engineer hiring process includes a round that does not exist in traditional software interviews: you are asked to design, in front of another person and in English, a system that uses a language model. An assistant over a company's documents, an agent that takes actions, a chat for a million users. You are not graded on transformer math. You are graded on how you think: whether you ask before you draw, whether you turn the answers into numbers, whether you know what the model decides and what the code decides, and above all whether you can **choose and say out loud what you give up** with each decision. This guide trains that round with eight original cases, ordered by how often they show up in real interviews. Each case takes place at a fictional company and follows the same seven-step script: clarify, set the boundary, sketch, name the trade-offs, evaluate and cost it, break it, and wrap up. The numbers (requests per second, tokens, cost per request, latency) come from short Python scripts that run without an API key. Each case ends with a complete interview written in English and annotated with a rubric, and every lesson gives you the exact phrases to lead that step in English. After the cases, two modules for the rounds people forget to prepare: the **forward deployed** interview, where you break down an ambiguous business problem and talk to a simulated client, and defending **a project you built** as if it were a take-home.