CommerceOS
Commerce Data Platform
portfolio.dev › projects › commerceos
A stateless, event-sourced commerce data platform processing 1M+ operations/day with GenAI-powered inventory forecasting.
Live deployment
Opens inside portfolio — no new tab
CommerceOS Architecture
System architecture overview
Ingress
- REST API (FastAPI)
- Pub/Sub Topics
Processing
- Event Consumers
- ETL Workers
- GenAI Forecasting
Storage
- BigQuery
- GCS
- Redis Cache
Resilience
- Circuit Breakers
- DLQ
- Retry Budgets
Data flows
- →API → Pub/Sub → Microservices → BigQuery
- →Inventory events → LLM forecast → Reorder signals
- →Failed events → Dead-letter queue → Replay
Overview
CommerceOS is a commerce data platform built on GCP using event-sourced microservices. Each service deploys independently, enabling fast iteration without coordinated releases. The platform ingests, transforms, and serves commerce data at production scale.
Architecture
Built on GCP with BigQuery for analytics, Pub/Sub for event streaming, Cloud Run for stateless services, and FastAPI for REST APIs. Docker containers with CI/CD pipelines enable independent deploy paths per microservice.
- Event-sourced microservices with independent deploy pipelines
- Pub/Sub message bus for async communication between services
- BigQuery as the analytical data warehouse
- Circuit breakers, retry budgets, and dead-letter queues for resilience
GenAI forecasting layer
Added an LLM-powered forecasting layer that surfaces reorder signals 48–72 hours ahead of stock depletion. This reduced stockout incidents by approximately 60%, turning reactive inventory management into proactive planning.
Impact
- Designed a stateless, event-sourced microservices system processing 1M+ operations/day with independent deploy pipelines.
- Added a GenAI forecasting layer surfacing reorder signals 48–72 hours ahead of depletion, reducing stockout incidents by ~60%.
- Implemented circuit breakers, retry budgets, and dead-letter queues, cutting downtime ~40% and improving response time ~25%.