Festival week triples kiosk turns at the ice rink. The conversational model is unchanged; people are waiting. What is the right move?
Select an answer to reveal the explanation.
Short Explanation
A longer line is a traffic problem, not a sudden vocabulary crisis. Scale out replicas of the booth you already validated. Save model experiments for quality, not for crowd size.
Full Explanation
When festival week triples kiosk turns at the ice rink and the conversational model is unchanged while people wait, the right move is to add replicas of the already-tested conversational app rather than starting a customization experiment. Civic festival load is a traffic problem: scale out the validated booth instead of treating queues as a sudden vocabulary crisis. Associate Helm/Kubernetes depth covers replica scaling for demand spikes so public visitors get capacity, not a fresh lab digression.
Launching a brand-new ASR vocabulary experiment before adding any capacity fails by confusing crowd size with model quality. Rewriting CUDA kernels to stretch one pod across the festival crowd fails by escaping into out-of-scope kernel work instead of horizontal scale. Replacing Riva with a VIA Blueprint because queues always mean the wrong modality fails by misreading load as a modality mismatch.
Exam caveat: load growth against a validated pipeline is replica scaling, not an automatic customization restart. Operational check: confirm model and chart are unchanged and healthy, increase replicas for the packaged conversational app, watch queue length and latency, and reopen model experiments only if quality—not capacity—regresses.