A fintech app calls a fraud scorer for every card swipe and must answer within 100 milliseconds at the 95th percentile, with traffic that spikes during evenings. Which hosting choice best fits?
Show answer and explanation
Answer: D
A Vertex AI online prediction endpoint keeps model replicas warm and answers synchronous requests with low latency, while autoscaling adds or removes replicas as traffic changes. That combination matches a per-swipe, latency-bound workload with evening spikes. Batch results arrive too late to gate an authorization, notebooks are not production services, and message-driven functions add asynchronous hops and cold-start delay.
Why the other options are wrong
- A. Batch scoring is economical for offline work, but results written to a table arrive after the swipe has already been decided, so they cannot gate the authorization.
- B. A notebook kernel is a development environment with no availability guarantees or autoscaling, and tunneling to it would not meet production latency or reliability needs.
- C. Pub/Sub adds an asynchronous hop, and a function started per message can incur cold-start latency, so the answer would not reliably return within the synchronous time budget.