Hypergrowth scaling made simple
summary
Argues that hypergrowth systems succeed by resisting novelty and keeping architecture minimal: Java 17 microservices on Kubernetes, ML in Python with a homegrown feature store, Google Cloud for infra, and Postgres for everything (even analytics and event storage), intentionally avoiding Kafka and document DBs. Presents scale evidence — 27 million customers and 500k businesses — supported by a small ops team of 15 ops engineers and only dev and prod environments. Claims this enables extreme speed, recommends deferring irrevocable decisions and optimizing for delivery rather than premature cleanliness.
AI-written summary · about
chapters (8)
- 0:00Introduction and goals
- 3:20System design evolution
- 13:01Defining system complexity
- 18:19Revolut's simplicity approach
- 24:25Managing complex systems
- 29:36Hyperscaling and legacy
- 37:02Summary and conclusions
- 43:41Q&A session
