\cleardoublepage \part{Foundations} \setbookpart{Foundations}
Part I: Foundations
An AI platform is not a model with an interface in front of it. It combines data products, model and prompt pipelines, retrieval stores, agent runtimes, gateways, compute and controls, and it has to behave predictably even though the components at its centre behave probabilistically. This part therefore establishes the vocabulary, requirements and reference structure on which the rest of the book relies.
Chapter 1 introduces the book, its place in the series and the two fictional organisations, Nordvik Region and Hansa Mutual, whose platforms are built and defended chapter by chapter. Chapter 2 recaps the six usage modes of large language models and what the platform must provide in each, because every mode moves the control point and opens a different attack surface. Chapter 3 condenses European regulation into engineering requirements, from logging and robustness to human oversight and data protection by design; the full regulatory map belongs to AI Governance. Chapter 4 presents a layered reference architecture with a cross-cutting control plane, and Chapter 5 shows how AI Decision Records, model cards and the architecture repository keep design decisions traceable.
What you will learn in this part:
- How this book divides the subject with AI Governance and Architecture as Code
- Where each usage mode puts the control point, and the baseline the platform owes each mode
- Which articles of the AI Act, the Cyber Resilience Act, NIS 2, DORA and the GDPR become platform capabilities, and which artefacts prove them
- A reference architecture with a control plane for identity, policy, observability and evidence
- How to record AI design decisions traceably
A team running AI in production can hold its estate against Chapters 3 and 4 to see what is missing; a team that has not started finds there its first platform.