So, with three strikes against it already, why am I still writing about it? Simply put - it's rather good. What it tries to do is to identify good quality websites first, and then particularly good pages after that. The idea is that you give it a key concept, and then other keywords to fine tune the search. It then goes off and number crunches for a while (minimum of an hour) and then throws back a whole heap of results. These results are based on site, then on pages in the site. Some of these pages are unlikely to turn up in the results you'd run with a 'normal' search, so while it's not exactly searching the deep web, it does bring you back pages you won't otherwise see.
It's also important to know that it's not giving you answers, just sites and pages that may contain the answer to your query, so to that extent it's also a little bit like directory search engines (remember those?) and so it's a two stage process.
It's a really interesting approach, but I think the aforementioned obstacles are going to cause big problems for it. However, if you have an hour to spare in order to let a search engine run around after you and then email you when it's done, give it a bash!