Databricks showcases real-time retail AI recommendation system

Databricks detailed how a top Asian fashion e-commerce platform uses its tools to build real-time product recommendations for over 1 million monthly users. The system ingests around 1,000 user actions per second—like product views and cart updates—directly into Databricks' lakehouse platform, then serves personalized suggestions without storing every interaction, reducing delays. The architecture streamlines both offline training and instant, in-session recommendations.

• System manages 1,000 user events per second
• All data flows through Databricks' lakehouse platform
• Uses direct user signal input for real-time recommendations
• Supports catalogs with 100,000+ items and 1M active users
• Reference setup omits separate message broker infrastructure