why we built a multi-model LLM router January 22, 2026 • #llm #litellm #vertex-ai #infrastructure #reliability Google's DSQ started throttling us mid-event