Our risk and operations teams watched trading activity on dashboards backed by a nightly warehouse load. Every incident review started with the same complaint: by the time the numbers showed something odd, the window to act had closed hours earlier. We needed per-symbol and per-account metrics that were seconds old, plus the ability to drill into individual trades, with queries fast enough for dashboards refreshing every few seconds. We built the pipeline on Kafka, Flink SQL and Apache Doris. Th...