Many AI-centric discussions lack the fundamental importance that the underlying architecture has for the business performance of the resulting system. At the example of vamos.ai, a media mining solution to extract metadata from videos with machine learning, this session introduces a powerful architecture pattern for AI-native business software: based on containerized micro-services and event-driven. Also the presentation shows the implementation of this architecture pattern called centros.one: an open sourced MLOps workbench and environment for scalable and reliable AI business applications. The case study gives specific insights, how the vamos.ai architecture drives collaboration between Data Scientists and DevOps teams, resulting in a fast and efficient way to integrate AI models from a variety of sources into a consistent overall system. Attendees will take away important inspirations around a modern, AI-native architecture pattern that can be re-used for a variety of other cases and learn how to build a media mining service.