Agent (AI agent). A software system that uses an AI model to interpret goals and select actions within an environment, such as calling tools, retrieving information, or initiating workflows, under defined authority.
Agentic AI. AI systems capable of pursuing goals through tools, state, reasoning, and action, beyond generating a single response.
Architecture decision record (ADR). A concise document recording a significant architectural decision, its context, alternatives, rationale, and consequences.
Attribute-based access control (ABAC). Authorization that evaluates attributes of the user, resource, action, and context, rather than role alone.
Authoritative source. The system or process designated as the trusted origin of particular information or attributes.
Batch processing. Processing data in groups at scheduled or triggered intervals.
Canonical model. A shared representation of important enterprise concepts to which local representations can be mapped.
Change data capture (CDC). A pattern that identifies inserts, updates, and deletes in a source system and propagates them downstream.
Chunking. Dividing content into retrieval units for indexing in a retrieval system.
Context engineering. The practice of selecting, structuring, and governing the information supplied to an AI system for a task.
Data catalog. A system that describes data assets and their metadata to support discovery, understanding, and governance.
Data contract. An explicit agreement between a data producer and consumers covering schema, semantics, quality, compatibility, and change.
Data lake. A large-scale repository that stores data in many formats, often in object storage, with structure applied when data is consumed.
Data lakehouse. An architectural approach that combines scalable, open storage with managed tables, transactions, governance, and support for analytical and AI workloads.
Data lineage. The record of where data originated, how it was transformed, and where it is used.
Data mesh. A domain-oriented organizational and architectural approach to data based on domain ownership, data as a product, self-service infrastructure, and federated governance.
Data observability. The ability to understand the condition and behavior of data systems and products through signals such as freshness, volume, schema, distribution, and lineage.
Data product. A governed, discoverable, reusable data capability designed for identified consumers, with an owner, contract, quality expectations, and lifecycle.
Data steward. A role responsible for the practical management of definitions, metadata, quality issues, and governance processes for a data domain.
Data warehouse. An analytical environment that integrates structured data from multiple sources into consistent models for reporting and analysis.
Decision intelligence. The architecture and capabilities that combine data, analytics, rules, models, and workflows to improve and govern enterprise decisions.
Domain. A coherent area of business responsibility and knowledge, often used as a boundary for data ownership.
EDA-IQ. Enterprise Data Architecture for Intelligence and Quality: the conceptual framework used in this book, comprising five connected dimensions — trusted data foundation, information and data product architecture, data platform architecture, semantic and knowledge architecture, and intelligence architecture.
Embedding. A numerical vector representation of content intended to capture aspects of its meaning, used for similarity search.
Enterprise intelligence. The organizational capability created when trusted information, technology, people, processes, and intelligent systems work together to improve decisions and actions.
Entity resolution. Determining whether different records refer to the same real-world entity.
Event-driven architecture. An architecture in which systems publish events describing business occurrences that other systems consume.
Federated governance. A governance model in which enterprise-wide rules are set centrally while domains manage domain-specific decisions within those rules.
FinOps. The practice of bringing financial accountability to variable cloud consumption.
Fine-tuning. Further training of a model on selected data to adapt its behavior for particular tasks.
Generative AI. AI systems that generate content, such as text, code, or images, in response to inputs.
Golden record. A trusted, consolidated representation of an entity; often more accurately achieved through attribute-level authority.
Guardrail. A boundary that defines what teams or systems may do, allowing local decisions within it.
Hallucination. A model output that is fluent but unsupported or incorrect.
Human-in-the-loop. A design in which a person reviews, approves, or decides before an AI-proposed action takes effect.
Hybrid retrieval. Retrieval that combines techniques such as keyword search, vector similarity, structured filters, and reranking.
Idempotency. The property that repeating an operation produces the same result, preventing duplicate effects on retry.
Information domain. A logical grouping of related enterprise information with meaningful business ownership.
Intelligent enterprise. An organization that operationalizes enterprise intelligence.
Knowledge graph. A representation of entities and the relationships among them, with properties and provenance.
Large language model (LLM). A model trained on large volumes of text that can interpret and generate natural language.
Least privilege. Granting each identity only the access required for its approved purpose.
Master data. Shared data describing core business entities such as customers, products, and suppliers.
Metadata. Data that describes other data, including technical, business, operational, and governance characteristics.
Model gateway. A controlled access layer between applications and AI models providing authentication, routing, logging, and policy enforcement.
Ontology. A formal representation of concepts within a domain and the relationships and constraints among them.
Operating model. The way an organization structures responsibilities, decision rights, processes, and funding to deliver its strategy.
Platform engineering. Building reusable internal capabilities that make delivery teams more productive and reliable.
Prompt injection. Content crafted to influence a model’s behavior in ways that conflict with the application’s intended instructions.
Pseudonymization. Replacing direct identifiers with substitutes while preserving some analytical utility; distinct from anonymization.
Reference architecture. A reusable model describing the capabilities, relationships, and principles required for a class of enterprise outcomes.
Reference data. Controlled codes and classifications, such as country or currency codes.
Reranking. A second-stage process that reorders retrieved candidates by relevance.
Responsible AI. Designing and operating AI systems with attention to reliability, transparency, oversight, fairness, accountability, and potential harm.
Retrieval-augmented generation (RAG). An architectural pattern in which an application retrieves information from external sources and supplies it to a generative model as context.
Schema evolution. Managing changes to data structures over time without breaking consumers.
Schema-on-read. Applying structure to data when it is consumed rather than when it is stored.
Semantic architecture. Architecture concerned with meaning, concepts, relationships, and business semantics.
Semantic layer. A governed mapping of business concepts and metrics to underlying data.
Service-level objective (SLO). A target level for a measured service indicator, such as freshness or availability.
Strangler pattern. Incrementally replacing a legacy system by building new capabilities around it and redirecting consumers over time.
Streaming. Processing data continuously as events occur.
Technical debt. The accumulated cost of design shortcuts and legacy structures that constrain future change.
Tokenization (data protection). Replacing sensitive values with substitutes that can be resolved only through a controlled service.
Trust engineering. Combining expectations, controls, accountability, and operational feedback so that consumers can judge whether data is suitable for a purpose.
Vector database. A system that stores vector representations and supports similarity search, often used within RAG architectures.