E-Commerce Data Platform
Olist Brazil — Real marketplace data
End-to-end data pipeline over 100,000+ real orders from Brazilian e-commerce. PostgreSQL + dbt with 4 model layers, RFM segmentation, cohort retention and revenue trends in pure SQL. CI/CD with GitHub Actions and a dashboard deployed to production.
100k+
orders analyzed
4
dbt layers
3
analytical marts
CI/CD
GitHub Actions
Pipeline architecture
The full flow from raw Kaggle data to the production dashboard.
Raw CSV Files
Kaggle · Olist Brazil Dataset · 9 relational tables
PostgreSQL
Source database · local Docker
dbt — Staging
stg_orders · stg_customers · stg_products · stg_payments · stg_sellers
dbt — Intermediate
int_orders_enriched · int_seller_performance · business logic
dbt — Marts
mart_customer_rfm · mart_cohort_retention · mart_revenue_trends
Live Dashboard
Next.js · Vercel · ecommerce-dashboard-puce.vercel.app
dbt model structure
Each layer has one clear responsibility. Staging has no joins; intermediate is never exposed to the dashboard.
Renaming, type casting, deduplication. Single source of truth.
Joins and business logic. Not exposed to the dashboard.
Final analytical tables. Consumed by the dashboard.
SQL — Key models
The project's three analytical marts. Click to expand the code.