Think what problem you are facing with the web right now and how you would solve it. Don't solve a problem for a thousand or a million people because all of them will have different problems. Start with you. Do you not like the interface? Do some css play, mockups on gimp or pen and paper.

What are your queries usually? Take a look at your own Google history, find and try to group what you are looking for and what would you ideally find that Google/other search engines don't satisfy.

I understand your desire to do something "revolutionary". But most things are incremental.