Difficulty: 8/10real-time scoring, feature store, model deploymentReal-time processing architecture with machine learning models, a rules engine, and a distributed database, detecting 95% of fraudulent transactionsScenario Questions0-2 years experience1How would you build a simple rule‑based service that flags a transaction if the amount exceeds a user’s daily limit?2If you receive a transaction event with a missing merchant category, what would your fraud detection pipeline do?3What data would you need to compute a basic risk score for a single transaction in real time?2-5 years experience1Follow-up QuestionsWhat would you monitor to detect model drift in production?How would you balance latency requirements with the need for richer features?Can you describe a fallback mechanism if the ML service is unavailable?Suggest Improvement
