Hello all, i want to locally host an end-to-end prototype of an Ai voice agent system (low scale). What should my tech stack be? preferably open source.

What do you usually track for call metrics? duration, sentiment, rate of booking vs inquiries (if applicable), if caller ended call prematurely. Do you think it is important to track if caller always interrupts the agent and doesn't let the agent finish talking? Am thinking that's a sign the agent is talking too much. But right now I'm having a hard time getting that part right - it's counting more interruptions than the number of times the agent spoke. What else do you track?

what are the metrics and benchmarks i should consider when choosing a llm for voice ai agent? for local development.

I'm doing some consulting work for a legal firm who has just done a successful proof of concept (PoC) using ElevenLabs for a voice agent. However, now we are running into some regulatory hurdles when it comes to things like data retention and data sovereignty. I'm wondering if anyone has experience with deploying Vapi within an enterprise environment? I'm not sure if full end to end on-prem self-hosting is possible, but I suspect it is if the org is on the Enterprise plan. I'm also curious to know whether, now the use case has been pretty clearly defined by the PoC, we would be better off building and deploying on prem using PipeCat etc? I haven't played around with it yet, but I can see that @Hugo Podworski is a fan, and I'm wondering if this makes more sense to use than a product like Vapi in an enterprise environment?

VIDEO LINK: https://youtu.be/wjeAYO6e4ac --- CODE: https://github.com/HugoPodworski/first-pipecat-agent --- If you have any questions about the code please just let me know in the comments