MLOps Mastery - Learning Notes

A structured learning path toward MLOps mastery, covering tools, platforms, and core concepts across the full ML lifecycle.

---

Concepts

Foundations

• ML lifecycle stages (data, training, evaluation, deployment, monitoring)
• Technical debt in ML systems
• Reproducibility and experiment tracking
• ML system design patterns
• Offline vs online evaluation

Data Engineering for ML

• Data versioning and lineage
• Feature engineering pipelines
• Feature stores (online vs offline)
• Data validation and schema enforcement
• Handling data drift and skew
• ETL/ELT for ML workloads

Model Training & Experimentation

• Experiment tracking and metadata management
• Hyperparameter optimization strategies
• Distributed training patterns
• GPU/TPU resource management
• Model registry and artifact management
• Transfer learning and fine-tuning workflows

Model Serving & Deployment

• Batch vs real-time inference
• Model serialization formats (ONNX, TorchScript, SavedModel)
• Blue/green and canary deployments for models
• Shadow deployments and A/B testing
• Model compression and quantization
• Edge deployment considerations

CI/CD for ML

• Continuous training pipelines
• ML-specific testing (data tests, model tests, integration tests)
• Automated retraining triggers
• Pipeline orchestration patterns
• Infrastructure as Code for ML

Monitoring & Observability

• Data drift detection
• Model performance degradation tracking
• Prediction logging and auditing
• Alerting strategies for ML systems
• Feedback loops and ground truth collection

Governance & Reliability

• Model explainability and interpretability
• Fairness and bias auditing
• Regulatory compliance (GDPR, model cards)
• Cost optimization for ML infrastructure
• Disaster recovery for ML systems

---

Tools & Platforms

Experiment Tracking & Model Registry

• MLflow
• Weights & Biases (W&B)
• Neptune.ai
• CometML

Data & Feature Management

• DVC (Data Version Control)
• Feast (Feature Store)
• Great Expectations (Data Validation)
• Delta Lake / Lakehouse formats
• Apache Spark for ML

Pipeline Orchestration

• Apache Airflow
• Kubeflow Pipelines
• Prefect
• Dagster
• ZenML
• Metaflow

Model Serving & Inference

• TensorFlow Serving
• TorchServe
• Triton Inference Server
• BentoML
• Seldon Core
• Ray Serve
• vLLM (LLM serving)

Containerization & Infrastructure

• Docker
• Kubernetes
• Helm charts for ML workloads
• Terraform / Pulumi
• KServe

CI/CD & Automation

• GitHub Actions for ML
• GitLab CI/CD for ML
• CML (Continuous Machine Learning)
• DVC Pipelines

Monitoring & Observability

• Evidently AI
• Prometheus + Grafana
• Arize AI
• WhyLabs

Cloud ML Platforms

• AWS SageMaker
• Google Vertex AI
• Azure Machine Learning
• Databricks MLflow

LLMOps (Large Language Models)

• LangChain / LangSmith
• LlamaIndex
• Prompt engineering and management
• RAG (Retrieval-Augmented Generation) pipelines
• Fine-tuning LLMs (LoRA, QLoRA)
• LLM evaluation frameworks (RAGAS, DeepEval)
• Guardrails and content filtering

---

Progress

Area	Status
Foundations	Not Started
Data Engineering for ML	Not Started
Model Training & Experimentation	Not Started
Model Serving & Deployment	Not Started
CI/CD for ML	Not Started
Monitoring & Observability	Not Started
Governance & Reliability	Not Started
Tools & Platforms	Not Started
LLMOps	Not Started