The Interview Question Map: AI Engineer, FDE and AI PM

You get the interview. Five minutes in, you're doing fine. Then the panel asks how you handled the ambiguous cases, or what you'd measure in the first two weeks, or what the launch criteria were, and you freeze, because your project never had to answer that question before.

This resource fixes that before it happens. It's a map of the exact questions every panel actually asks about an AI project, organised by the three role tracks, AI Engineer, Forward Deployed Engineer and AI PM, with the specific artefact that answers each one and a model answer built from a real reference build.

Inside:

How an AI project interview actually goes. The five-minute walkthrough you control, and the ten-plus minutes of interrogation you don't, where the panel picks one claim and pulls on it until it holds or it doesn't.

The AI Engineer question map. Three questions, tied to three artefacts, the architecture decision record, the eval harness, and the repo and README, each with a model answer: why structured output over free text, how you handled the ambiguous cases, and what the eval harness actually looks like.

The FDE question map. Three questions tied to the discovery doc, the deployment plan and the reliability section: how you'd deploy inside a real customer's workflow, what you'd measure in the first two weeks, and how you'd handle a team that doesn't trust the system yet.

The AI PM question map. Three questions tied to the PRD, the metrics plan and the launch and rollout plan: what the launch criteria are, what metric tells you it's working, and how you'd roll out to one user first.

Ten questions that come up regardless of role. The judgment questions every track gets asked, from "why this problem" to "what breaks it," each one worth a two-sentence answer prepared before you walk in, not during.