How would you design robust, fault-tolerant systems in GCP for large-scale AI workloads at Character.AI?
