I am building a low-latency, real-time voice agent using AssemblyAI's WebSocket API and Gemini for asynchronous LLM routing.