Funny thing. I'm currently building one right now - disturbingly close to your product idea. It's currently a "generalized feed consumer" that presents a user with "stream of [de-duplicated] links to consume..."
I'm just starting to use it for most of my content consumption, so that I have a nice corpus of data about my preferences. The goal is to eventually develop a predictive model to start filtering content with...
It's an idea I've kicked around for a long time too. Building a simple Bayesian system would be neat; as you read, you should be able to tag items easily, and in a way that doesn't pigeonhole you into only seeing one side of an argument (I'd like to still see the opposition's view in, say, a political discourse for instance).
I've got too much else to work on currently, but I'd love to see what you come up with :).
I'm just starting to use it for most of my content consumption, so that I have a nice corpus of data about my preferences. The goal is to eventually develop a predictive model to start filtering content with...