Walk through the architecture of a distributed caching layer designed specifically for storing KV caches in multi-turn LLM conversations.
