Decompose: here are 8,000 taxi trips. Propose a product, sketch the components and APIs, buildable in a week.

What separates answers here is whether you notice the survivorship problem unprompted. The file contains completed trips, so every passenger who waited and gave up is invisible, which quietly rules out the demand-prediction product most candidates propose in the first two minutes. Saying 'this dataset cannot see unmet demand, so I will not build a product that depends on it' is the single highest-scoring sentence available in this round. The one-week constraint is doing work too: it is there to see whether you can pick one user and one decision instead of describing a platform.