@Vapi great blog article https://vapi.ai/blog/how-we-solved-latency-at-vapi. Is that shenanigan also applied when using our own openai keys ?