Smart home assistant

Speculative decisions first, LLM fallback only when needed.

TypeSafe's reference demo evaluates smart-home requests with multiple questions in parallel and escalates uncertain cases to an LLM.