How Medium Powers Real-Time Recommendations at 1M OPS

What this post added

This post details Medium's migration from a relational features data model to a list-based approach within their ScyllaDB-powered feature store to support real-time recommendations at 1M OPS. It explains the limitations of the previous relational model for handling time-series interaction data and introduces the new list-based schema in ScyllaDB, which stores lists of relation IDs with timestamps. This change enables more efficient querying and retrieval of user interaction history, crucial for personalized recommendations.