Following on from my other voice idea, it would be kind of fun to make a context table. That is, a device that listens to your conversation, pulls out keywords and searches for them, displaying a kind of passive live-updating screen with information relevant to the context. You'd need to do a bit of processing to figure out what words tend to be searchworthy (proper nouns, stuff outside the top N most common words etc), but it wouldn't have to be terribly accurate to work.
In fact, most of what you'd need to build a simple web version is just the Web Speech API generating AJAX requests to Wikipedia's API.
Obviously, the coolest thing would be if you could then turn that into a table that automatically displays relevant things to what you're talking about but, hey, gotta start somewhere.