Semantic Layer

AI governance encompasses the policies, processes, and technical controls organizations apply when using AI systems, including data classification access control, manual review of model outputs, security scanning of AI-generated code, token cost quota management, and compliance audit logs.

The semantic layer is an abstraction layer situated between raw data storage and business applications, transforming technical database structures into business-friendly conceptual models and defining unified calculation logic for business metrics.

语义层描述约4KB即可包含核心业务指标定义、维度层级关系、业务规则和字段别名映射等信息

构建结构化语义层比模型微调或复杂Prompt工程成本更低、可维护性更强，且对不同模型都能产生正向效果

If no one on the team can maintain the semantic layer, forcing in a heavy-governance tool will lead to model rot, and people will eventually bypass it and pull numbers themselves, wasting the governance investment.

Governance capability is the invisible dividing line for BI tools: a semantic layer like LookML can fundamentally solve the "which number is right" problem, at the cost of requiring people who know SQL to maintain the models.

Almost every AI+BI stack has a semantic layer (answering "what does revenue mean"), yet completely lacks a canonicalization layer (answering "was this same question already answered ten minutes ago in different wording").

A semantic layer can ensure consistent metric definitions, but it cannot stop the data warehouse from re-executing a full scan and aggregation because the query wording changed.

The fix is to intercept Agent-generated queries before they reach the data warehouse and rewrite them onto the semantic layer's metric definitions, making metric calls easier to cache than raw SQL.

Semantic grounding binds the metrics in a user's question (such as "profit margin" or "target achievement rate") to the organization's approved business definitions and calculation logic, often relying on a metrics layer or semantic layer to achieve this; only when both schema grounding and semantic grounding are correct can the SQL be both structurally correct and business-correct.

A semantic layer is an abstraction layer between the data warehouse and upper-layer applications, defining the query interface in business terms rather than physical table columns, with benefits including governance, reuse, and decoupling.

In essence, the semantic layer establishes a translation dictionary between raw data tables and final queries, encapsulating SQL fragments as business terminology.

The semantic layer pre-defines metrics, dimensions, and business rules, giving AI a clear frame of reference when parsing natural language.

The semantic layer has existed within the OLAP (Online Analytical Processing) ecosystem since the early days of BI tool development.

Databricks Genie introduces the concept of a semantic layer, which reduces the probability of the model generating incorrect queries by pre-defining business metric definitions and relationships between tables.

Traditional semantic layers face structured queries (SQL) with clear rule boundaries, whereas LLMs receive natural language intent, whose boundaries are ambiguous and dynamically changing.

The concept of a semantic layer has existed in the data analytics field for many years, typically used to define business metrics, dimensions, and measures so that different tools have a unified understanding of concepts.

The Semantic Layer establishes a mapping between raw database fields and business language; without it, Agents can produce inconsistent results due to variations in how questions are phrased.

Using an LLM shifts developers from the role of 'thinking about how to implement' to the role of 'reviewing and patching AI output'.

dbt, Looker Semantic Layer, and Cube.js are the mainstream choices enterprises currently use to build semantic layers.