True Real-Time Personalization with Vectors

This session provides a practical exploration of building a true real-time personalization engine using Kafka and Flink, showcased through the lens of one of the largest online fashion retailers. Key topics covered include:

• Ingesting CRM and stock system data through Change Data Capture.
• Incorporating live event data from Rudderstack.
• Utilizing Flink to seamlessly join and build models.
• Creating vectors for effective data representation.
• Implementing a recommender system.
• Exposing results through an API for immediate consumer interaction.

In contrast to traditional systems that claim real-time capabilities but typically operate on segmented data from the night before, our approach executes these processes within seconds. This ensures genuine real-time personalization, responding promptly to changes in products or user behaviour. The implementation has not only transformed the speed of personalization but has also resulted in an 11% increase in revenue compared to existing systems.