Hi, I am working on building a conversational AI agent pipeline having my own custom STT, TTS, and LLM components. I am looking to train and deploy my custom models for this project. For this, I am exploring LiveKit for the first time, but stuck on how to proceed in implementing these components without using plugins or inference. Can someone please help me regarding this?

If your LLM supports the OpenAI api you can just use the OpenAI plugin for that and point it at the inferenec service. For TTS/STT you have a look at one of the plugins that is most like yours and use that as a starting point. See: https://github.com/livekit/agents/tree/main/livekit-plugins