Scaling ML Pipelines for Financial Exception Management

In this session I walked through the architectural challenges of detecting anomalies across millions of financial transactions — and the transition from legacy Oracle PL/SQL procedures to a modern, distributed PySpark architecture.

Key takeaways

• Reducing false positives by 40% using Isolation Forests.
• Handling data skew in distributed systems without blowing the latency budget.
• Why “human-in-the-loop” is non-negotiable for financial AI.