Skip to content
Ayan Putatunda
Back to work

Semantic layer on Snowflake

production

Zendesk Data Platform, 2025–2026

A unified business vocabulary across 10+ domains, built with dbt metrics and Snowflake Cortex semantic views, so people and agents ask questions in the same words and get the same answer.

Snowflake CortexCortex Analystdbt metricsSemantic viewsDynamic SQL

Problem

Every team defined revenue, active account, and ticket volume slightly differently. Analysts spent their week answering ad-hoc requests that only existed because there was no shared definition to point to, and any LLM put on top of the warehouse inherited the disagreement.

Approach

  • Modelled one governed vocabulary in dbt metrics, then exposed it through Snowflake Cortex semantic views so natural-language questions resolve against the same definitions as dashboards.
  • Cortex Analyst for AI-assisted SQL generation, bounded by the semantic layer rather than raw tables.

Impact

  • Ad-hoc query requests down 45%.
  • Self-serve analytics opened to non-technical stakeholders.
  • Governed natural-language analytics over petabyte-scale data.