which stack (Stt + llm + tts) are you guys using, that does decent in latency and intelligence