Implementing LLM Router Architecture: Dynamically Selecting Models by Cost, Latency, and Task Complexity in Production | Wicked Smart Data