To support 6 million on-demand rides per day, a lot has to happen in near-real time. Latency translates into missed rides and monetary losses. Grab relies data streaming in Apache Kafka, with Scylla to tie it all together. This presentation details how Grab uses Scylla as a high throughput, low-latency aggregation store to combine multiple Kafka streams in near real-time, highlighting impressive characteristics of Scylla and how it fared against other databases in Grab’s exhaustive evaluations.