Serverless ML: Lessons from Capital One

Capital One migrated 2,000 applications to AWS a decade ago and now utilizes a serverless-first approach for over a third of its apps, including an anomaly detection ML model for transaction monitoring. The benefits of migrating an ML model to serverless architecture include simplified provisioning, improved scaling, and reduced maintenance requirements. Challenges in the migration process included data size, memory limitations, and runtime constraints, which were addressed through solutions such as data sampling, library layers, and code optimization.