Present a Self-Built AI Project and the Product Decisions It Required

In a product manager interview at Microsoft, you are asked whether you have built anything on your own time, for example through "vibe coding" (making software mostly by describing what you want to an AI tool and letting it write the code), that shows how you approach products built around AI. Answer with an example. ```hint One build, your decisions Pick a single project and prepare to explain the choices only you could have made: why a model belonged at its core, what you did when its output was wrong, and what changed after someone else used it. Tools and buzzwords carry little weight on their own. ``` ### Clarifying Questions - Does a project count if only I used it, or are you looking for something other people used? - Should I go deep on one build, or briefly cover a few? - Are you more interested in AI as part of what the product does, or in AI as the way I wrote the code? ### What a Strong Answer Covers - A specific build with a real user problem, and a reason an AI-based approach fit it better than a conventional one - Product decisions that only arise with AI: wrong or inconsistent output, evaluation, cost and latency, user trust and control - Evidence of iteration based on what users or test data showed - Candid reflection on the limits of AI-written code and how the candidate verified it - A credible link from the build to how the candidate would work on AI features as a Microsoft PM ### Follow-up Questions - How did you decide the model's output was good enough to rely on? What would you measure if a million people used it? - What did the AI tool get wrong while you were building, and how did you find out? - If a feature you owned at Microsoft added a generative AI capability, which lessons from this project would you apply first?