I have a requirement to make dialog systems, but it’s desirable for us to make all inference within our own VPC and not have to make any external api calls, say to RASA servers to get inference. Of course the application does talk to the internet but only with one isolated service. We’re not allowed to make external HTTP calls to any other destinations. Is it possible to deploy constructed models & make inference entirely within an isolated VPC?
Welcome to the forum! 
I’m not sure I understand. If you don’t want to make API calls then just don’t write actions that do that. You can even build a chatbot completely offline.
Yes, of course. What is the problem?