I was really hoping for a bit more meat in this article. Perhaps some discussion on their caching strategy, or how they're building their JSON (i.e. traditional model-based #to_json approach, or something along the lines of Rabl).