How would you design a robust, fault-tolerant system for serving large language models in the cloud for Character.AI?
