I had an idea for this where I was building a manually-constructed index of sites kind of like old-school Yahoo (related to typography), and then there would be a full-text search of those URLs and their close relatives (pretty much, the idea being that for a given URL, I'd index anything else in the same URL directory or subdirectories of that URL). I ended up giving up on it largely because time is finite and I have plenty of other projects competing for my time. I imagine something that built on the basic idea though could work, by omitting the Yahoo-style directory and instead working from a collection of vetted URL starting points.