In the middle of considering a swap from 11labs convAi to vapi. Vapi has considerably more options available to us, but quite sure that 95% of them will be likely useless. I want hoping to find a guide (ideally official one by vapi) on the de facto setup for conversation agent, mostly focusing on performance (ie reduced latency). I find it strange that it marks the google LLMs as higher latency to (thinking) OpenAI models, my experience doesn't tend to agree? Anyway, any tips on the best configuration for performance would be much appreciated. TIA