Discuss feature engineering from transaction data, model architecture for real-time scoring, handling concept drift, the feedback loop problem with fraud labels, and precision-recall trade-offs.

Discuss the trade-off between catching fraud (recall) and avoiding false positives that block legitimate transactions (precision). Explain how the cost ratio depends on the business context.

Define 'active user,' discuss feature engineering from early user behavior, model selection, evaluation strategy, and how this model would be used to optimize onboarding.

Discuss behavioral biometrics, device fingerprinting, session-level features, and how to build a model that distinguishes legitimate users from attackers without excessive friction.

Given transactions and accounts tables, define 'suspicious' criteria (e.g., sudden spikes, unusual geographies, rapid velocity), and write an efficient query using aggregations and window functions.

Use window functions to compute rolling metrics, join with merchant metadata, and handle edge cases around new merchants and low-volume categories.

Walk through a structured root cause analysis. Segment by platform, geography, payment method, merchant category. Check for product changes, bugs, and external factors.

Frame this as a causal inference problem. Discuss how to isolate the effect of the policy from other factors, potential quasi-experimental approaches, and what metrics capture 'trust.'

Combine user transaction data, market size estimates, competitive landscape, and user research to build a prioritization framework. Discuss how you'd validate your recommendations.

Define primary and guardrail metrics, choose the randomization unit, handle the merchant-buyer two-sided dynamic, and discuss how to measure long-term impact on retention.