dbt + MetricFlow + MCP
Interactive technical case study

One definition.
Many consumers.

Explore how a governed semantic layer turns raw GA4 activity into business metrics that humans, dashboards and AI agents can use without silently redefining what the numbers mean.

This is not a dashboard.It is a visual explanation of the decisions encoded in metrics.yml and semantic_models.yml, and how those decisions travel through an AI-facing analytics stack.
01 / Metrics

What does the number actually mean?

Select a metric to see its formula, grain, assumptions, trade-offs and failure modes. Featured metrics expose the full governance text from the semantic layer.

02 / Trace

Follow one question through the stack

Instead of asking an LLM to invent SQL, watch the system discover an approved metric, read its meaning, find valid dimensions and then delegate the calculation.

Example question

“What’s the activation rate for mobile users?”

03 / Semantic layer

How does the system know what can be grouped by what?

The semantic model is the contract between dbt tables and business metrics. Select a metric to highlight the model that owns its measures, then inspect its entities, dimensions and measures.

Semantic model map
4 models, connected by user and session entities
Why this mattersA dimension is not merely a column that happens to exist. The semantic layer declares which concepts belong to which analytical grain, so consumers can ask legitimate questions without recreating join logic themselves.