Content, media operations and enterprise workflows for a national streaming service, on one backend.
Editors publish and expect the change live before they finish the sentence. Readers arrive in bursts of tens of thousands and expect the page from the closest, dumbest layer available. The previous system served both badly: minutes to publish, and a database that fell over on launch nights.
Cached the resolved content graph, not pages — keyed by entity and version, so a publish makes the old key unreachable rather than wrong.
Moved media ingestion to TUS resumable uploads feeding RabbitMQ workers, so transcoding runs unattended and survives the API restarting.
Split playback analytics into ClickHouse; PostgreSQL keeps everything with a lifecycle.
Publish-to-live went from minutes to seconds. Launch-night load stopped being an event. The platform now also carries CRM and billing, including the SAP BRIM integration.