Interview OS
Back to AI / ML

Library

MLOps & serving

Train, deploy, monitor.

Systems

Overview

MLOps is the discipline of shipping and maintaining models in production: versioning data and models, CI/CD, monitoring for drift, and safe rollouts. It's what turns a notebook experiment into a reliable service.

How it works

Systems
TrainRegistryServeMonitorPipelineversionedModel storeregistryDeployAPIDrift
ClientDataServiceEdge

Step by step, with examples

  1. 1

    Pipeline

    • Reproducible, versioned training.
  2. 2

    Model store

    • Track versions and lineage.
  3. 3

    Deploy

    • Batch or online endpoints.
  4. 4

    Drift

    • Detect drift and trigger retrain.
    • Example: canary rollout

Overview

Operationalize models: feature stores, CI/CD for models, batch vs online serving, monitoring for drift.

Common pitfalls

  • Train/serve skew
  • No drift monitoring
  • Ignoring rollback paths

Where this content comes from

For full transparency, this content is curated and verified from these sources:

Peer-reviewed ML papers & model cardsProduction ML engineering referencesOppZen-authored ML eval rubrics