I've developed a website / Firefox toolbar that allows people to surf the web by topic or within specific domains. I think it has huge potential and that I'm on to something big here, but its usefulness relies on having a powerful recommendation engine behind it, something I cannot build myself. It's similar in many ways to stumbleupon (at least in regards to how the rec engine needs to operate. new content is introduced by users who fave websites from their toolbar and should be introduced to more and more users as it gains popularity within a period of time).
I've talked to a few recommendation engine providers (loomia, strands, etc). Strands is the only company that may be able to provide a back-end recommendation service that could work for my service (great API, content agnostic, ability to filter by item tags).
But still, I'm not sure a 3rd party service like this could ever provide the flexibility that I would eventually want (For instance, how could I control the rate at which new content is spread and eventually stops being recommended? how would I address potential issues of gaming?).
I'm not that knowledgeable about how rec engines operate. If anyone here understands the technical details regarding rec engines (or has experience building them) I would love to hear what you think. A third party system could potentially allow me to launch this thing very soon and would alleviate many scaling issues. But maybe it would be smarter to try to get a team together and build it from scratch so that I have complete flexibility.
It doesn't get into any technical implementation details but does cover design, strategy, and issues you should consider.
You might want to browse that chapter at your favorite bookstore.
Also check out "Programming Collective Intelligence" by Oreilly. I haven't read it yet, but it is sitting on my shelf.