Applied MLOps: Machine Learning Model Deployment & Monitoring
The 87% of models that never reach production fail on the last mile — packaging, serving, and monitoring. This track takes a trained model to a monitored, auto-retraining production service: containerised serving, CI/CD, a feature store, and drift detection with automated rollback. Built around what SRE and ML platform teams actually run.
What You'll Learn
What's Included
Interactive learning path with AI-generated lessons read aloud by your personal tutor
Step-by-step lab instructions with AI guidance — run in your own IDE or sandbox.
AI interviewer asks tailored questions, gives real-time feedback and a score report
MCQ questions (theory, practical, code) to test your knowledge and earn a badge
Compete head-to-head against AI on real production problems — judged on correctness, performance, scalability, code quality, and edge cases
Course Syllabus
Module 1: MLOps Foundations
- •Why 87% of models never ship — the last-mile problem
- •The ML lifecycle vs the software lifecycle
- •MLOps maturity levels (manual → CI/CD → fully automated)
- •Reproducibility: data, code, and environment versioning
- •Experiment tracking and a model registry with MLflow
Module 2: Packaging & Serving Models
- •Serialisation: pickle, ONNX, and framework formats
- •Wrapping a model in a FastAPI inference service
- •Containerising with Docker and slim images
- •Batch vs real-time vs streaming inference patterns
- •Load testing and latency/throughput trade-offs
Module 3: CI/CD & the Feature Store
- •Automated testing for ML: data, model, and integration tests
- •Building a CI/CD pipeline (build → test → deploy) for a model
- •Model deployment strategies: blue-green, canary, shadow
- •Feature stores and preventing training/serving skew
- •Rollback and safe promotion between environments
Module 4: Monitoring & Observability
- •Operational metrics: latency, error rate, saturation
- •Data drift vs concept drift and how to measure them
- •Statistical drift detection (PSI, KS-test) on live features
- •Prediction and outcome logging for ground-truth join
- •Dashboards and alerting on model quality decay
Module 5: Automated Retraining & Governance
- •Trigger-based and scheduled retraining pipelines
- •Champion/challenger evaluation and automated promotion
- •Model lineage, audit trails, and reproducible releases
- •Cost, GPU utilisation, and scaling considerations
- •Capstone: closed-loop pipeline with drift-triggered retraining
You're all set
Everything in this course is unlocked.
Continue Learning →