NYC Taxi Analytics Platform

Why I built it

A dataset large enough that the loading strategy actually matters. Reloading it in full every run is the obvious approach and the wrong one, so this build is structured around only moving new data.

Questions it answers

• What does trip demand look like across time, zone and vendor?
• How do you keep a high-volume table current without a full reload each run?
• Where should business logic live — in the report, or in the model underneath it?

How I built it

1. Metadata-driven pipelines. Table configuration drives the ingestion, so the pipeline count does not grow with the table count.
2. Incremental loading. Only new records move each run, which is what makes the platform viable at this volume.
3. SQL Warehouse layer. Cleaned data lands in a warehouse where the model is defined in SQL rather than assembled in the report.
4. Semantic model and DAX. Measures live in the semantic model so every report reads the same definitions.
5. Report design. The Power BI report went through five iterations to land on a navy and yellow palette with an active-filters bar, so users can always see what they are looking at.

How I know it works

• Incremental runs load only changed data — verified against warehouse row counts.
• Measures are defined once in the semantic model, not duplicated per visual.
• The workspace was migrated intact, which proved the build was not tied to one environment.