Hi Sofiaya,
Here is a video from Embedded World 2021 that covers a tinyML project using FreeRTOS https://www.youtube.com/watch?v=YJjxjozCJjA.
If you wanted to interact with bigger models, it would probably make more sense (and only be feasible) to interact with it over the internet. You could use coreHTTP to make HTTP requests to a server hosting a bigger model for example. This video shows how you can setup your own model in AWS bedrock and interact with it through invoking a lambda from API gateway https://www.youtube.com/watch?v=aMAmD-1SFYQ. You could of course interact with any consumer AI that has an HTTP API.
Hope this helps.