pipecat-ai/pipecat ? reverse-engineered prompt

I want to build a backend for a real-time AI voice assistant. The core idea is to create a Python service that can handle a smooth, streaming conversation with a person.

It should be able to take in audio from a user's microphone, convert that speech to text, send it to an LLM to figure out a response, and then use a text-to-speech service to talk back to the user. The most important thing is that it's super fast and has very low latency, so it feels like a natural conversation.

Make it flexible so I can easily plug in different AI services for transcription, intelligence, and voice generation. It should also be built to handle more than just audio in the future, like video streams. Set up the basic Python server for this conversational pipeline. A simple example to test it would be great.