Senior machine learning engineer in Honolulu, HI. I build reliable systems from unreliable AI.
What I learned about prompt caching on the Anthropic API — cost savings, invalidation gotchas, and architecture patterns that actually stick.