While it's not even close to true AI, I'm working on an app that adds about 1000 levels of intelligence to my car for my connected car project called "bluemonkey". He doesn't do any learning, which is, imho, a good thing, except what i teach him via code. The last thing I want is to be guilty of writing skynet.
He basically controls everything that goes on in the car and exposes a bit of that through the Internet Chat Relay protocol so I can tell him to do stuff and also get information from him. He's currently idling in test form on the #linuxice channel on freenode.
Anyway, i like the idea you are going for. Hope to see how this works out for you.