Recorded Walkthrough of a Past AI Project and Your Own Role in It

Record a short video in which you walk through an AI project you worked on in the past: what the project is and does, and which parts of it you were personally responsible for. This is the first item of a timed online assessment for a forward-deployed engineering role focused on AI agents, and 10 minutes are allotted to it. It works as a brief project and behavioral screen, not a coding task. ```hint Pick the project for the role Choose the project where you can speak in the most detail about decisions you made yourself, and where you can say who used it and what changed for them. ``` ```hint Make your part unmistakable A reviewer watching the video cannot ask follow-up questions. Plan the exact sentences that separate what you built and decided from what teammates, libraries or hosted models provided. ``` ### Constraints and Clarifications - The item is recorded, not live: nobody can interrupt to ask what you meant, so the recording itself has to answer the questions a live interviewer would ask. - The 10 minutes are the time allotted to the whole item. How much of it is preparation, and whether the video has a minimum or maximum length, is not stated; check the on-screen instructions before you start. ### Clarifying Questions Because there is no interviewer, settle these from the instructions, or decide them yourself, before recording: - Do the 10 minutes include preparation time, and is there a length limit for the video itself? - Can you share your screen to show a diagram, a demo or code, or is it a camera-only recording? - Must the project come from professional work, or does a side project or academic project qualify? - Is there a single take, or can you re-record? ### What a Strong Answer Covers - A one-sentence statement of the problem, who had it, and why an AI approach fit - A clear account of the system: data, model or LLM usage, how it was evaluated, and how it was deployed or used - Explicit ownership: what you personally designed, built and decided, separated from the work of teammates and vendors - Evidence of outcome: a measured result or concrete user feedback, plus one honest limitation or failure and what you learned from it - A planned structure that finishes within the time, with emphasis on what matters for deploying AI systems for customers ### Follow-up Questions - What was the hardest technical decision in the project, and which alternative did you reject? - How did you know the model's output was good enough, and what did you do about the cases where it was wrong? - If a customer wanted this project running inside their own environment next month, what would you change first? - What would you do differently if you started the project again today?