Just an observation I've made, but isn't this almost identical to what WolframAlpha's goals are? With the exception of handling your own personal data, they both seem to work in the same ways: Understanding natural language and interpreting massive amounts of data in different ways. I'd be interested to see how they are different underneath.