Skip to main content

Complete MLOps: industrializing the machine learning lifecycle

Take a machine learning model from an exploratory notebook to a production system that retrains itself, monitors itself, and rolls back when it drifts.

Course Duration: 10h

What You'll Learn

  • Diagnose the gap between a notebook that works and a service that runs
  • Pin every source of non-determinism so a run is reproducible six months later
  • Track experiments, version data and models, and promote them through a registry
  • Package a model as a container and ship it through a continuous delivery pipeline
  • Choose between batch, online, and streaming serving based on latency requirements
  • Detect data drift and concept drift with statistical tests and useful alerts
  • Trigger automated retraining and roll back a bad model in a single command

Prerequisites

  • Supervised learning (Course 04)
  • Feature engineering (Course 06)
  • Basic Docker: images, containers, environment variables

Course Modules

  1. What MLOps solves that classic software development ignores
  2. Reproducibility: seeds, environments, version pinning
  3. Experiment tracking with MLflow
  4. Versioning data and models
  5. Model registry and promotion stages
  6. Containerizing a model
  7. Continuous integration and delivery pipelines
  8. Batch, online, and streaming serving
  9. Monitoring: data and concept drift
  10. Automated retraining and rollback

The running example

A single project traverses the ten modules: a subscriber churn prediction model for a telecom operator. Module 1 opens it as a Jupyter notebook full of implicit assumptions. Each subsequent module adds a brick — reproducibility, tracking, versioning, a registry, a container, a pipeline, a service, monitoring, retraining — until module 10 delivers a system that retrains on drift, validates before promotion, and rolls back in one command.

Tools you will use

MLflow for experiment tracking and the model registry. DVC for data and model versioning against a remote store. Docker to freeze the runtime. GitHub Actions for CI/CD. FastAPI for the online service (revisited in Course 40). Evidently for drift reports. Every tool is introduced when the running project needs it, never in isolation.

Assessment and certificate

The course ends with a 40-question exam covering every module. On success, a certificate of completion is issued; its number is verifiable on the platform.

Free courses, by contrast, end with a 5-question quiz and a preview of the certificate, without certification.