Introduction to MLOps

Overview:

Build Python foundations for automation, reproducibility, and experiment orchestration in ML pipelines.

Topics to Cover:

• Python foundations for MLOps tasks (i.e., packaging, CLI creation, unit testing)
• Create reproducible code for automation and deployment
• Streamline experiment tracking and orchestration with Python tools

Overview:

Apply data engineering practices to manage versioning, features, and pipeline integration for ML workflows.

Topics to Cover:

• DE practices for ML pipelines (i.e., DVC, data versioning, workflow orchestration)
• Process structured/unstructured data & engineer features with PySpark
• Work with embeddings, feature stores, and pipeline integration

Overview:

Train and manage models with Ray, SageMaker, and MLflow to support production-ready pipelines.

Topics to Cover:

• Train models with Ray, SageMaker, and MLflow
• Best practices: managing models in production pipelines

Overview:

Package models with Pickle/ONNX to ensure portability, reproducibility, and deployment compatibility.

Topics to Cover:

• Serialize & package models using Pickle & ONNX for efficient deployment across platforms and environments
• Ensure model portability, reproducibility, and compatibility for production use

Overview:

Deploy models locally or in the cloud with serverless tools and CI/CD automation.

Topics to Cover:

• Deploy models locally and in the cloud
• Use AWS Lambda and serverless tools to reduce infrastructure overhead
• Automate CI/CD pipelines to ensure model freshness and reliability

Overview:

Track metrics, detect drift, and automate retraining with MLflow, Evidently, Prometheus, and Grafana.

Topics to Cover:

• Monitor metrics like accuracy, latency, and drift using MLflow, Evidently, Prometheus, and Grafana
• Detect data and concept drift and visualize changes to ensure long-term model reliability and fairness
• Set up alerting pipelines and automate retraining when performance issues or anomalies are detected

Overview:

Contrast traditional MLOps with LLMOps, covering RAG pipelines, prompt engineering, and evaluation shifts.

Topics to Cover:

• Compare core differences between traditional MLOps and workflows for large language models (LLMs)
• Explore prompt engineering, RAG pipelines, and new infrastructure patterns
• Understand evaluation shifts from accuracy to groundedness and human feedback