Knowledge Base

Enterprise Data & AI Glossary

Plain-language, vendor-neutral definitions of the terms that shape modern enterprise data and AI work — from lakehouse architectures and data mesh operating models to SAP modernisation, retrieval-augmented generation, and the EU AI Act. Each entry is anchor-linkable, so you can cite or share a single definition directly.

Agentic AI

Agentic AI refers to systems in which large language models plan, reason, and execute multi-step tasks autonomously by calling tools, querying data, and reacting to intermediate results rather than returning a single response. An agentic system typically combines a reasoning model with tool interfaces, memory, and an orchestration loop, and enterprise deployments add guardrails, evaluation, and human-in-the-loop checkpoints for consequential actions.

Apache Kafka

Apache Kafka is an open-source distributed event streaming platform that stores streams of records in durable, partitioned, append-only logs replicated across a cluster of brokers. Producers publish events and consumers read them independently at their own pace, which makes Kafka a common backbone for real-time pipelines, change data capture delivery, and event-driven architectures.

CDC (Change Data Capture)

Change data capture (CDC) is a set of techniques for detecting inserts, updates, and deletes in a source system — most reliably by reading the database transaction log — and propagating those changes downstream in near real time. CDC enables low-latency replication and event-driven pipelines without repeated full extracts, reducing load on source systems and keeping analytical copies current.

Data Contract

A data contract is a formal, versioned agreement between a data producer and its consumers that specifies schema, semantics, quality expectations, ownership, and service levels for a dataset. Contracts are most effective when enforced automatically in CI/CD and pipeline runtime, so breaking changes are caught before they reach downstream consumers.

Data Fabric

A data fabric is an architectural approach that uses active metadata, knowledge graphs, and automation to provide a unified layer for integrating, governing, and accessing data across distributed sources without physically centralising it. Where data mesh is primarily an organisational operating model, data fabric is primarily a technology pattern, and the two are often combined in practice.

Data Governance

Data governance is the framework of policies, roles, standards, and processes that ensures data is available, understandable, high quality, secure, and used in compliance with regulation. It assigns accountability through ownership and stewardship models, and modern implementations increasingly express policies as code so they can be enforced automatically.

Data Lakehouse

A data lakehouse is an architecture that combines the low-cost, open-format object storage of a data lake with the management features of a data warehouse, including ACID transactions, schema enforcement, and performant SQL. It is typically implemented with open table formats such as Delta Lake, Apache Iceberg, or Apache Hudi, allowing BI, data engineering, and machine learning workloads to run on a single copy of the data.

Data Lineage

Data lineage is the recorded trace of where data originates and how it is transformed as it moves through systems, from source to report or model. Lineage supports impact analysis before changes, root-cause analysis during incidents, and evidence for regulatory and audit requirements, and it can be captured at table, column, or job level.

Data Mesh

Data mesh is a sociotechnical approach to analytical data management that decentralises ownership to business domains instead of a single central data team. It rests on four principles: domain-oriented ownership, data as a product, a self-serve data platform, and federated computational governance. Data mesh addresses organisational scaling bottlenecks; it is an operating model rather than a specific technology stack.

Data Observability

Data observability is the continuous, automated monitoring of data health across dimensions such as freshness, volume, schema changes, distribution, and lineage. Its goal is to detect anomalies and pipeline failures before downstream consumers are affected, borrowing practices from software observability and applying them to datasets and pipelines.

Data Quality SLO

A data quality SLO (service level objective) is a measurable target for a data quality dimension — such as freshness, completeness, or accuracy — agreed between a data producer and its consumers. It adapts site reliability engineering practice to data: underlying indicators (SLIs) are monitored continuously, and breaches or exhausted error budgets trigger defined remediation.

DataOps

DataOps applies agile, DevOps, and lean manufacturing practices to data and analytics delivery: version control, automated testing, CI/CD, orchestration, and monitoring for data pipelines. The aim is shorter cycle time from raw data to trusted insight, with fewer defects, through collaboration between data engineers, analysts, and platform teams.

Delta Lake

Delta Lake is an open-source storage framework, hosted by the Linux Foundation, that adds a transaction log on top of Parquet files to bring ACID transactions, schema enforcement and evolution, time travel, and upserts to data lakes. It is one of the principal open table formats used to implement lakehouse architectures, alongside Apache Iceberg and Apache Hudi.

Digital Twin

A digital twin is a virtual representation of a physical asset, process, or system that is continuously synchronised with operational and sensor data from its real-world counterpart. Twins are used to monitor condition, simulate scenarios, and predict behaviour — for example forecasting equipment failure or testing process changes before applying them physically.

EU AI Act

The EU AI Act is the European Union's regulation on artificial intelligence, adopted in 2024 with obligations phasing in over subsequent years. It takes a risk-based approach — prohibiting certain practices, imposing strict requirements on high-risk systems, and setting transparency duties for limited-risk uses and general-purpose AI models — and it applies extraterritorially to providers placing AI systems on the EU market.

Federated Governance

Federated governance is a model in which a central function defines global policies, standards, and interoperability rules while domain teams implement and enforce them locally, close to the data. In data mesh it appears as federated computational governance, where policies are automated and embedded in the platform rather than enforced through manual review boards.

FinOps

FinOps is an operating model and cultural practice that brings financial accountability to variable cloud spend by making engineering, finance, and business teams jointly responsible for cost decisions. It is commonly described through iterative inform, optimise, and operate phases, covering visibility and allocation, rate and usage optimisation, and continuous governance of spend.

Knowledge Graph

A knowledge graph represents entities and the typed relationships between them as a graph, usually with an ontology or schema that gives the data formal semantics. Knowledge graphs support reasoning, semantic search, and data integration across silos, and they are increasingly used to ground large language models with explicit, verifiable facts.

LLM Orchestration

LLM orchestration is the coordination of large language model calls with prompts, tools, retrieval, memory, and control flow to accomplish tasks that a single model call cannot. Orchestration layers handle routing between models, chaining of steps, error handling, and observability, and they are the backbone of production RAG pipelines and agentic systems.

Medallion Architecture

Medallion architecture is a layered data design pattern, common in lakehouses, that organises data into Bronze, Silver, and Gold tiers. Bronze holds raw ingested data, Silver holds cleansed and conformed data, and Gold holds business-level aggregates ready for analytics and reporting. Each hop incrementally improves data quality and structure while preserving raw history for reprocessing.

RAG (Retrieval-Augmented Generation)

Retrieval-augmented generation (RAG) is an architecture in which relevant documents are retrieved — typically via vector or hybrid search — and injected into a large language model's context before it generates an answer. RAG grounds model output in current, proprietary, or domain-specific data without retraining, reduces hallucination, and enables citation of sources.

SAP BW Bridge

SAP BW Bridge is a transition option that embeds BW 7.5-compatible modelling — including extractors, data flows, and ABAP-based transformations — inside SAP Datasphere, so organisations can reuse existing BW investments while migrating to the cloud in stages. It carries known limitations relative to classic BW: BEx queries and analysis authorizations are not brought across, and reporting consumption happens through Datasphere rather than in the bridge itself.

SAP Datasphere

SAP Datasphere is SAP's cloud data management service on the SAP Business Technology Platform and the successor to SAP Data Warehouse Cloud. It combines data warehousing, cataloguing, semantic modelling, data integration, and federation, with the goal of keeping business context and semantics attached to SAP and non-SAP data as it is shared across the enterprise.

SAP S/4HANA

SAP S/4HANA is SAP's ERP suite built natively on the SAP HANA in-memory database, featuring a simplified data model, the Fiori user experience, and embedded analytics. It is the successor to SAP ECC, which faces announced end-of-mainstream-maintenance deadlines, making S/4HANA migration and the analytics continuity around it a major enterprise transformation topic.

Semantic Layer

A semantic layer is an abstraction that maps physical data structures to business-friendly entities, metrics, and dimensions with consistent, governed definitions. It lets BI tools, applications, and AI systems query the same certified metrics rather than re-deriving logic per tool, which reduces metric drift and is increasingly used to ground AI answers in trusted business definitions.

Vector Database

A vector database stores and indexes high-dimensional embedding vectors and answers similarity queries using approximate nearest neighbour algorithms such as HNSW or IVF. Because embeddings place semantically similar content close together in vector space, vector databases power semantic search, recommendations, and the retrieval step of retrieval-augmented generation.