Data Architecture for AI

Most AI systems don't fail because of the model. They fail because of what comes before it. Wrong data. Missing context. Undefined ownership. Stale knowledge. Silent contract violations. By the time the model generates a wrong answer, the real problem happened several architectural layers earlier. The model didn't cause the failure — it exposed it. Data Architecture for AI is a practitioner's guide for engineers and architects who need to build the data foundation that makes intelligent systems actually work — not just in demos, but in production, at scale, under real-world constraints. Across 7 parts and 25 chapters, this book builds a complete architectural framework: Part I establishes the central argument: AI does not create data problems, it makes existing ones impossible to ignore. You will learn how to audit your data estate for AI readiness and identify the technical debt that will surface the moment a language model touches your data. Part II introduces the Three Surfaces — the conceptual core of the book. Facts, relationships, and context are not the same thing and cannot be stored as if they were. Every chapter that follows is a development of this principle. Part III covers temporal knowledge: how to model change rather than snapshots, implement bi-temporal data patterns, and ensure your AI system knows what it knew — and when. Part IV examines what a language model actually sees in your data: the vocabulary problem, the null value problem, the schema pattern decisions that determine whether retrieval produces correct answers or plausible-sounding ones. Part V addresses contracts, governance, and ownership — the organisational architecture of data. Including how to design data contracts that survive schema evolution, detect silent violations before they reach the AI layer, and define ownership boundaries that hold under production pressure. Part VI builds the knowledge layer: semantic layers, knowledge graphs, and real-time streaming architectures for keeping knowledge bases current without full rebuilds. Part VII closes with evaluation frameworks, data observability patterns, migration playbooks for legacy systems, and ten Architecture Autopsies — real failure patterns with root cause analysis and the correct design that would have prevented each one. Every chapter includes AWS-specific implementation guidance covering Aurora, Neptune, Glue Data Catalog, Lake Formation, DataZone, Bedrock Knowledge Bases, Kinesis, and CloudWatch. This is not a book about prompting. It is not a model comparison guide. It does not cover agent orchestration, inference optimization, or certification preparation. It is written for the data engineer stepping into AI architecture, the solutions architect designing production AI platforms, and the senior engineer who has already watched a well-engineered RAG pipeline fail — and needs to understand why. Reliable AI systems are not built by choosing a better model. They are built by designing data, knowledge, and system boundaries correctly. This book shows you how.