The role Stream Processing plays for an Opti Channel Experience
summary
Bank implements a 5‑second SLA event pipeline using Apache Kafka, Apache Beam feeding Flink, HDFS for history and CDC via Informatica to expose personalized nudges over WhatsApp; a reinforcement multi-armed bandit picks nudges from four runtime inputs out of 30+ data sources and updates on clicks (1/0/−1). HBase/Cassandra were rejected, system lands on a fast KV store, took ~6–9 months to stitch together, and production runs with limited triggers so far.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:27Jamies background
- 0:54Agenda
- 1:27Disclaimers
- 4:14Requirements
- 6:06Multichannel and omnichannel
- 7:34Old architecture
- 8:03New architecture
- 10:30Multiunbanded
- 11:30Personas
- 13:00Wrap up
- 13:27Brand control
- 13:46State
related talks
34mTransforming the financial services industry with Event Stream Data
42mBuilding a Streaming (E-health) Data Pipeline: When and How?
42mDistributed Real-Time Stream Processing: Why and How
25mData: Moving with the flow
25mSharing what we love: The Apache Flink story at Pinterest
47mReal-time Fraud Detection Challenges and Solutions
30mSmart Streams: A Real-time framework for scoring Big and Fast Data
46mFast Data at ING