MLOps
Operationalize machine learning: lifecycle, versioning, pipelines, experiment tracking, serving, Kubernetes and drift monitoring.
What Is MLOps?
The discipline of operationalizing machine learning: people, pipelines, and platforms.
The ML Lifecycle
From problem definition to retirement: every stage a model passes through.
Reproducibility & Versioning
Version code, data and models together — the foundation of trustworthy ML.
Data Pipelines
Reliable, repeatable data flow: ingestion, validation, cleaning and feature preparation.
Feature Stores
One source of truth for features: consistent online and offline features.
Experiment Tracking
Log every run: metrics, hyperparameters and artifacts with MLflow.
Model Registry
Stage models through lifecycles and keep production versions audit-ready.
Training at Scale
Move from notebooks to managed training jobs with reproducible configs.
Model Packaging & Serialization
Wrap models into portable artifacts: pickle, ONNX, and MLflow models.
Model Serving APIs
Expose models over HTTP with FastAPI and MLflow serving — with proper request validation.
Containerization with Docker
Package the model, code and dependencies into one portable image.
Kubernetes Basics for ML
Orchestrate containers: pods, deployments, services and autoscaling.
Model Deployment Strategies
Ship models safely: shadow, canary and blue-green deployments with rollbacks.
Monitoring & Drift Detection
Watch predictions and features: data drift, concept drift and model performance decay.
Evaluation in Production
Estimate quality without labels: sampling, human review, and proxy metrics.
CI/CD for Machine Learning
Automate testing, training and deployment in pipelines that run on every change.
Testing ML Systems
Test data, features, models and infrastructure — because ML code is only part of the system.
Data & Model Governance
Auditability, access control and compliance for ML assets.
Cost & Performance Optimization
Balance accuracy against inference cost, latency and carbon.
LLMOps
Operationalize generative AI: prompt management, evals, guardrails and LLM monitoring.
MLOps Roadmap
Synthesize the course into a plan: build a production ML system and keep improving it.