A fintech built a real-time fraud model on DXData in 9 weeks

Lumen Pay runs fraud detection on every card authorization. They needed sub-second scoring with replayable history. They had neither.

Nine weeks later, their streaming ingest, feature store, and model scoring live on a single DXData catalog with one on-call rotation across three engineers. The old stack is decommissioned.

We used to argue about which team owned the Kafka cluster. Now we argue about which feature to add next.