08 — Implementations
Guiding question: “How is this cross-cutting executable pattern or end-to-end system implemented in code?”
This layer contains cross-layer, reusable executable reference architectures: system patterns that integrate multiple components across layers, and complete end-to-end examples that span data ingestion through deployment.
What belongs here:
- Production system patterns that integrate ≥2 layers (e.g. MLflow + feature store + serving)
- End-to-end reference architectures adaptable to real projects
- Reusable orchestration and deployment patterns
What does NOT belong here:
- Concept-specific implementation notes — those live with their parent concept. For example:
- XGBoost / tree ensemble code → 03 — Tree-Based Models
- SHAP / interpretability workflow → 03 — Evaluation and Model Selection
- Neural network PyTorch implementations → 03 — MLP and Representation Learning
- Transformer fine-tuning pipeline → 06 — Fine-tuning
- Business problem framing → 07 — Applications
- Theoretical derivations → 01 — Foundations
- Project-specific work → 09 — Projects
Placement rule: Implementation notes live with the concept unless they are cross-cutting reference architectures that integrate multiple layers.
01 — System Patterns
Reusable production patterns for ML/AI systems: experiment tracking, feature stores, model serving, monitoring, RAG pipelines, agents, quantization, CI/CD, and more. Each pattern integrates components across 05_ml_engineering/ and 06_ai_engineering/.
02 — End-to-End Examples
Complete ML/AI system walkthroughs combining multiple components across multiple layers. Each is a reference architecture spanning data → model → deployment → monitoring, adaptable to real projects.