Real-Time Fraud Detection in Java with Kafka, Streams & Vector Similarity

Ship a fraud detector without training a custom model. We’ll build a Java streaming pipeline that scores card transactions in near-real time using Kafka, Spring, behavior embeddings, and rule guardrails. You’ll see how to set thresholds, manage drift, and keep p95 latency low, focusing on explainability and operations rather than black-box ML.

Outline:

• Problem & constraints
• Architecture: Kafka → Spring services → vector similarity
• Feature/embedding design & thresholds
• Rules, explanations, and alerting
• Drift, replay, and observability
• Demo + Q&A