How was the caching layer organized, and what optimization features were used for your specific case?