GlobalScart is a multi-seller e-commerce marketplace that I built to simulate a production-grade digital commerce platform.

The platform allows customers to browse products, compare different seller offers, add products to cart, and place orders. Multiple sellers can sell the same product, and I implemented an algorithmic Buy Box that selects an eligible seller based on factors such as landed price, delivery speed, seller rating, and inventory availability.

For the backend, I used FastAPI with PostgreSQL as the transactional source of truth. I implemented seller-specific inventory with reservation and row-level locking to prevent overselling during concurrent orders.

For real-time event processing, I used a transactional outbox with Redpanda, and events are consumed by downstream systems such as ClickHouse. I also implemented real-time order-status updates using SSE and Redis.

For the data engineering side, I built an Airflow, PySpark, Parquet, BigQuery and dbt pipeline for historical analytics, while ClickHouse handles the realtime analytics layer.

One important part I implemented was zero-trust checkout. The backend doesn't trust prices or totals sent by the client. During checkout it revalidates the seller offer, price, shipping, seller eligibility and inventory, then creates an immutable order-price snapshot.

I also added authentication, RBAC, idempotency, rate limiting, reconciliation, DLQ handling, observability and automated testing. The current implementation has 410 automated tests passing and 43 database migrations verified.