Lessons

Concept pages that answer the customer-facing question — when do I choose this, what does it cost, and how do I explain it? — organized by the four-layer map. Where the labs teach by doing, lessons teach by deciding.

🎯 L1 · Apps & Agents

The application layer — where most engineering work sits. These pair with Labs 01–03.

• RAG Patterns — naive → hybrid → rerank, and when to go further
• Agent Architectures — hub-and-spoke, and why not a swarm
• Context Engineering — the bigger lever than prompt wording
• MCP and A2A — the protocols that connect agents to tools and each other

🏛️ L4 · Architecture & Governance

• Reference Architectures — pilot, agentic, and platform tiers, and which one matches a customer's stage
• Guardrails and Governance — NIST AI RMF, the EU AI Act, and ISO 42001 are a method, a law, and a certification — not the same thing

🗂️ L3 · Classic MLOps & Data

• The MLOps↔LLMOps Bridge — LLMOps extends MLOps, it doesn't replace it

🔜 Later layers

• L2 · LLMOps & infra — serving, quantization, vector DBs, gateways, observability (Phase 3+)
• L3 · Classic MLOps & data — the MLOps↔LLMOps bridge (Phase 4, in progress)

🔗 Links

• The Four-Layer Map — the structure these follow
• Labs — the hands-on counterpart
• Decision Frames — the build-vs-buy calls these inform