ML Systems

Building production-ready machine learning systems.

Overview

This skill category covers the complete ML system lifecycle:

1. Foundations – Core concepts, architectures, paradigms
2. Data Engineering – Data collection, quality, feature engineering
3. Model Development – Training, evaluation, frameworks
4. Performance – Optimization, acceleration, efficiency
5. Deployment – Serving, edge deployment, scaling
6. Operations – MLOps, monitoring, reliability

Categories

Foundations

• ml-systems-fundamentals – Core ML systems concepts
• deep-learning-primer – Deep learning foundations
• dnn-architectures – Neural network architectures
• deployment-paradigms – Deployment patterns

Data Engineering

• data-engineering – Data pipelines and quality
• training-data – Training data management
• feature-engineering – Feature creation and stores

Model Development

• ml-workflow – ML development workflow
• model-development – Model training and selection
• ml-frameworks – Framework best practices

Performance

• efficient-ai – Efficiency techniques
• model-optimization – Quantization, pruning, distillation
• ai-accelerators – Hardware acceleration

Deployment

• model-deployment – Production deployment
• inference-optimization – Inference optimization
• edge-deployment – Edge and mobile deployment

Operations

• mlops – ML operations and lifecycle
• robust-ai – Reliability and robustness

Key Principles

1. Data-Centric AI – Focus on data quality over model complexity
2. Iterative Development – Start simple, iterate based on metrics
3. Production-First – Design for deployment from the start
4. Monitoring – Continuous monitoring and improvement
5. Reproducibility – Version everything (data, code, models)

References

• Harvard CS 329S: Machine Learning Systems Design
• Designing Machine Learning Systems by Chip Huyen
• MLOps: Continuous Delivery and Automation Pipelines