Deploying CNN interview question

Had an interview question about how to deploy a CNN. Resume had no mention of CNN, DL or any tool kits like PyTorch, TF, etc. Background is in streaming (Flink, Spark Streaming) where I deploy ML models from MLLib in Spark Streaming and work with JVM Pojo objects / binaries in Flink. After I open up the ports and ID paths, write the apis to send / receive requests, write pipelines for the data to be sourced from Kafka queues, wire transmissions using grpc, load balancing through Nginx, service manager as Zookeeper, Groovy my way through build pipelines, artifact deployment by way of resoucing from Artifactory, all in an multi-cluster, on-prem environment. When I explained this to the interviewer, the only feedback was there is a disconnect, no other response. I don't think the interviewer was concerned about QPS or other metrics since we didn't get that far. I feel like he may have been looking for an obvious answer, but can't quite identify what he was maybe looking for online. Can someone shed light on what maybe was being asked and high level what would be a suitable response? Want to learn more, should be studying DL Ops now I guess because I feel like I've got a handle on ML Ops, but I'm drawing a blank on the DL side after this interview. #datascience #DL #dataengineerinterview