Receive daily AI-curated summaries of engineering articles from top tech companies worldwide.
Endigest AI Core Summary
This post explains how Estée Lauder Companies used Cloud Run worker pools to build a scalable, pull-based infrastructure for their consumer-facing AI applications.
•Estée Lauder migrated their Rostrum LLM platform to a producer-consumer model using Cloud Run worker pools and Cloud Pub/Sub to handle holiday-scale traffic for Jo Malone London's AI Scent Advisor.
•The web tier (FastAPI on Cloud Run Service) publishes user messages instantly to Pub/Sub, while worker pool instances act as always-on consumers handling LLM inference.
•The decoupled architecture delivered 100% message durability, strong UI latency SLAs, and minimal operations overhead without server management.
•Cloud Run worker pools support non-HTTP protocols via L4 ingress, NVIDIA L4 and RTX PRO 6000 GPUs, and are approximately 40% cheaper than request-driven services for long-running tasks.
•CREMA (Cloud Run External Metrics Autoscaler), an open-source tool built on KEDA scalers, enables automatic scaling of worker po
This summary was automatically generated by AI based on the original article and may not be fully accurate.
Google Cloud