MLOps interview questions and answers

The builder stage can contain compilers and test deps

Prefer safer formats such as ONNX, safetensors, or joblib only inside trusted images, and treat pickle as a last resort.

Liveness restarts a stuck process

Requests help the scheduler place Pods

The Horizontal Pod Autoscaler adds replicas from CPU, memory, or custom metrics such as queue depth or QPS.

Use a Job or a workflow CRD such as Kubeflow or Argo, not a long-lived Deployment.

Staging is for candidates that passed offline gates and maybe shadow traffic.

Log the entire Pipeline with mlflow.sklearn.log_model so preprocessing and estimator stay together, plus a conda or pip environment.

The tracking DB stores metadata

Operational monitoring is RED or USE metrics: rate, errors, duration, CPU, memory.

Request rate, p95/p99 latency, error rate, model version tag, feature freshness, PSI for top features, and prediction score histograms.

Each online feature should carry an event timestamp or ingestion time.

Hash a stable user id with a salt into buckets so the same user always sees the same model.

Besides the primary profit or default metric, guard approval rate, fairness slices, latency, and operational loss.

Bandits help when the cost of exploring a bad model is high and you want to shift traffic toward winners faster.

Producers drop feature keys on Kafka or SQS, workers pull, fetch online features, score, and write results back.

Use one registered artifact and the same feature definitions

The server waits a few milliseconds to group incoming requests into one GPU or SIMD forward pass.

Use Airflow, Prefect, Dagster, or similar to run extract, validate, transform, and materialize with retries and SLAs.

A contract is a schema plus semantics: types, null rates, allowed ranges, and owners.