ShareChat is a social media app with ~180 MAU and 50M DAU. We capture and aggregate various engagement metrics, viz. likes, views, shares, comments, etc., at a post level to curate better content for our users. In terms of numbers for the engagement metrics, we have writes and reads happening at a scale of 55k-60k ops/sec and 290k-300k ops/sec, respectively. With these engagement metrics directly impacting users, we need a datastore that would offer lower latencies and is highly available, resilient, and scalable. It would be better if we could achieve all of these at an optimal cost. This is to learn how we accomplished the abovementioned criteria by using in-house Kafka streams and ScyllaDB.