Independent project · Data engineering
E-commerce analytics platform
A cloud-native pipeline and dashboard that turns a 52 GB event dataset into interactive, sub-second business queries.
The problem
Raw event volume made interactive analysis impractical.
The source data contained more than 385 million e-commerce events. Business questions about product behavior, sessions, and conversion needed a model and query path designed for repeated exploration—not full-table scans.
My contribution
I built the data model, API, and dashboard independently.
The three-tier platform pairs a React frontend with a FastAPI service and a PostgreSQL star schema on Google Cloud SQL. The backend runs on Cloud Run; the dashboard provides the interactive view of the resulting measures.
- Partitioned event data by month so queries scan a smaller slice of the 52 GB dataset.
- Added materialized views for repeated aggregates and B-tree indexes on common lookup fields.
- Connected the data layer to an API and live dashboard rather than presenting query benchmarks alone.
Design decision & result
Optimize the questions people actually ask.
Monthly partitioning reduced scan size by 85%, while precomputed views and indexes supported sub-second queries. The project reports a 300–600× improvement over its baseline queries. Those figures describe the project benchmark, not a general guarantee for every workload.