16. Semantic Layers & Feature stores, Drew Banin, dbt Labs

• Semantic layers are in

  • Main idea: define your dataset and metrics, map out their relationships, translate semantic query into SQL
  • Example metrics: revenue per country, churn rate …

• One example

  

  • in dbt:

  

• Precision & consistency

  • Many people, many teams but only one way to define revenue
  • Avoid repeating work or copy-paste, and inconsistency can arise

• Bridging the gap

  • Standardisation: feature store for ML training and serving
  • Semantic layers: feature store in the BI world, output for analytics
  • Doing it once and correctly

  

  • get reuse and consistency