MLOps Fundamentals
MLOps applies DevOps principles to machine learning systems. It provides practices and tools for reliably building, deploying, monitoring, and retraining ML models in production so they remain accurate and maintainable over time.
itArtificial intelligence and machine learning | OpenSkills.info
Course pathWalk it in order
Look it upDip in anytime
Go furtherLeaves this page
Intro
MLOps Fundamentals
MLOps is the discipline of putting machine learning models into production and keeping them there reliably. A model that works in a notebook is not a product. MLOps bridges the gap between experimentation and operation.
Why MLOps exists
Machine learning systems fail in ways traditional software does not. The code may be correct, but the model degrades because the world changes. Input data drifts. Upstream schemas break. A model trained on last year's behavior gives wrong answers this year.
Traditional software engineering handles these problems with CI/CD, testing, monitoring, and rollback. MLOps adapts those practices for the specific challenges of ML:
- Models depend on data, not just code.
- Training is expensive and non-deterministic.
- Performance degrades silently without explicit monitoring.
- Reproducing a result requires tracking code, data, configuration, and environment together.
The ML lifecycle
An ML system moves through stages:
Continue the course
This section is part of the paid course.
See pricing to subscribe, or log in if you already have access.
Where this skill leads
Relevant careers
See how this topic contributes to broader role-level skill maps.
Sources
- https://docs.cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning
Supports
- MLOps maturity levels 0, 1, and 2 definitions
- Continuous training (CT) as distinct from CI/CD
- Automated pipeline architecture for ML systems
- Data and model validation as pipeline steps
- Monitoring and retraining triggers
- https://ml-ops.org/content/mlops-principles
Supports
- MLOps lifecycle phases and core principles
- Relationship between DevOps, DataOps, and ModelOps
- Feature store role in preventing training-serving skew
- Data drift as a cause of silent model degradation
- Reproducibility requirements for ML experiments
- https://learn.microsoft.com/en-us/azure/architecture/ai-ml/guide/mlops-maturity-model
Supports
- Five-level maturity model for ML operations
- Progressive deployment strategies (canary, shadow, A/B)
- Governance and audit requirements for production ML
- Latency and SLA considerations in model deployment
- https://mlflow.org/docs/latest/
Supports
- Experiment tracking (parameters, metrics, artifacts)
- Model registry for versioned artifact management
- Reproducibility through environment and dependency tracking
- Integration with popular ML frameworks
- https://aws.amazon.com/what-is/mlops/
Supports
- MLOps as adaptation of DevOps for ML workloads
- Pipeline orchestration and automation patterns
- Model monitoring and operational health tracking