Define your data once.
Trust it everywhere.
The problem
Three tools. Three answers. Zero trust.
Data Products
Raw tables become trusted Data Products
A Data Product is a governed dataset bundled with its business definitions, access policies, ownership, and lineage, treated like a product, not a one-off query.
- Discoverable in a shared catalog; any team can find and trust it
- Policy travels with the product: Role-based Access Control (RBAC), column masking, row-level security
- Versioned and tested: breaking changes reviewed before rollout
- Consumed by BI tools, APIs, and AI agents: same definition, every time
Data Products as code
Your business definitions. In git. In CI/CD.
Data Products can now be defined in YAML. Commit to git, test before merge, roll back when needed.
apiVersion: v1
kind: DataProduct
metadata:
name: Customer Lifetime Value
catalogName: analytics_catalog
schemaName: comprehensive_schema
dataDomainName: Enterprise Analytics Domain
summary: Lifetime value metrics per customer, refreshed weekly.
businessContext: |
- avg_order_value is the mean of sales_summary rows, not
of orders. That grain is one row per customer, day,
product and category, so it reads lower than a true
per-order average. Do not report it as AOV.
description: Revenue, order volume and tenure for every customer.
owners:
- name: Alice Johnson
email: alice.johnson@example.com
relevantLinks:
- label: Documentation
url: https://docs.example.com
tags:
- analyticsVersion control: Commit Data Product definitions to git alongside application code. Diff, review, merge.
Policy as code: Access rules live in the same repo as the definition. No separate admin console to remember.
CI/CD lineage checks: Validate upstream freshness automatically before publishing a new version.
The semantic layer
Your AI has no context. Data Products fix that.
The semantic layer translates raw data into business-ready context: metrics, dimensions, and relationships your tools and AI agents can trust. Data Products are that context layer, built into the query platform itself.
LLMs struggle with raw database schemas. They excel when given structured business context. Context, gained through the semantic layer, is what closes the gap between AI potential and AI accuracy.Starburst, What is the Semantic Layer?
Governed data products · Versioned · AI-ready