Data Academy · Layer 7 · Contextualize

Semantic Data Layer

Business objects and virtual semantic models; physical storage becomes an implementation detail.

Lesson 9 of 17 · Find it on the platform map

01 What it is

What this layer does

The semantic layer presents data as business objects — customer, order, invoice — with their relationships, metrics and KPIs, so users and applications ask for business concepts rather than tables. In SCIKIQ these are virtual semantic models and virtual tables, which makes physical storage an implementation detail. The layer is served through REST and GraphQL endpoints and supports multi-hop traversal, multi-object query and natural language to SQL.

02 Concepts

Four ideas to hold on to

1

Business object

A named business concept, such as customer or order, with its attributes and identity defined once. Consumers work with the object, not with the tables underneath.

2

Virtual semantic model

A model that defines objects, relationships and measures over source data without copying it. The definition is reused by every consumer.

3

Metric and KPI model

A single governed definition of how a measure is calculated, at what grain and with which dimensions. It stops the same KPI being calculated differently in each report.

4

Natural language to SQL

Translating a business question into a query. It is far more reliable when it targets a semantic model with defined objects and relationships than when it targets raw tables.

03 In SCIKIQ

What the layer contains

Business objectsSemantic models and virtual tablesMetrics and KPI modelsLogical relationshipsREST and GraphQL endpointsMulti-hop traversal, multi-object query and natural language to SQL

04 What good looks like

Signs it is working

  • Each headline KPI has one definition, and every report and agent uses it.
  • Applications read business objects through REST or GraphQL endpoints rather than querying source tables directly.
  • A change to physical storage can be absorbed in the semantic model without changing consumers.
  • Natural language questions return queries that can be inspected and that use the defined objects and relationships.

05 Diagnostics

Questions to ask your team

  1. 1

    How many definitions of revenue do we have in use today?

  2. 2

    If we moved a source table, how many reports and applications would break?

  3. 3

    Do our AI assistants query business objects or raw tables?

  4. 4

    Who owns each business object and approves changes to its definition?

06 Keep going

Related reading

— Questions

Frequently asked

Is a semantic layer the same as a BI model?

A BI model usually serves one tool. A semantic layer serves many consumers — reports, applications and agents — through shared definitions and endpoints.

What does multi-hop traversal mean here?

It means following relationships across several business objects in one query, such as from customer to order to product. It saves consumers from writing the joins themselves.

Do virtual tables copy data?

No. Virtual tables are defined over source data, so the semantic model can be changed without building and maintaining a new copy.