MLOps & Production ML
Models are easy. Systems are hard.
The engineering of keeping models working in production — pipelines, monitoring, versioning, and the discipline that separates demos from products.
Advanced3 modules · 10 lessons~91 min940 XPAligned with Google Production ML SystemsStart with: ML FoundationsStart with: Compute & Systems for AI
Progress0/10
Module
The ML Lifecycle
Decay, pipelines, and versioning
Models Rot on the Shelf9 min · 90 XPModels Rot on the Shelf
Pipelines, Not Notebooks8 min · 80 XPPipelines, Not Notebooks
Version Everything, or It Never Happened9 min · 90 XPVersion Everything, or It Never Happened
Module
Deployment Patterns
Serving topologies, safe rollouts, and tests that include the model
Batch, Real-Time, and the Space Between8 min · 80 XPBatch, Real-Time, and the Space Between
Ship It Like You Don't Trust It9 min · 100 XPShip It Like You Don't Trust It
The Skew That Kills Quietly10 min · 110 XPThe Skew That Kills Quietly
CI/CD for ML: Test the Model, Not Just the Code10 min · 100 XPCI/CD for ML: Test the Model, Not Just the Code
Module
Monitoring & Maintenance
The signals, the triggers, and the postmortem loop
Watch the Inputs, the Outputs, and the Money9 min · 90 XPWatch the Inputs, the Outputs, and the Money
Retrain on Evidence, Not on Vibes9 min · 90 XPRetrain on Evidence, Not on Vibes
Incidents, Rollbacks, and the Retro Loop10 min · 110 XPIncidents, Rollbacks, and the Retro Loop
Capstone project
Ship a model
Take any small model you've trained and put it behind an endpoint with versioning — even a simple FastAPI + Docker setup counts. Track one metric after 'deployment.'
Deliverable: A repo link with the serving code, a README on how to run it, and one monitoring decision explained.
Milestones
Finish every lesson to earn the MLOps Engineer Certificate.