Full talk · edited transcript

Production Patterns for Multi-Tenant AI: Scaling to 800 Banks on Databricks

Natasha Ueberschlag, Simon Schmitz18 Jun 2026

Simon Schmitz (senior data scientist, Atruvia) and Natasha Ueberschlag (manager of AI forward-deployed engineering, Databricks). Session at.

Serving 800 German banks on one AI platform required maintaining strict data isolation while scaling under banking regulations. The talk presents two production patterns that Atruvia implemented over the last year together with Databricks: a multi-tenant architecture for retrieval that deletes most of the usual infrastructure, and hierarchical retrieval with the RAPTOR algorithm for questions about a whole knowledge base. Around two-thirds of the audience had built generative AI applications, and fewer had built them in a multi-tenant environment.

Continue reading — it's free

Google or a one-time email code. No password.