The build your own local hosted smart speaker project has been a learning experience. Was using a 3B LLM for a conversation agent and while it worked, it was finicky. Currently trying a 7B model (the limit my hardware can handle) to see if its less finicky.