I don't know of any existing implementation, but I'd like to propose an alternative architecture. Assume that users (readers) have access to much larger stream

I don't know of any existing implementation, but I'd like to propose an alternative architecture. Assume that users (readers) have access to much larger streams of content, from heterogeneous sources, than they have attention for. And that each user's stream is unique in content and may overlap with those of other users differently for every other user. So the editor of a personalised web site provides not multiple alternative content but a single filter which can be applied to each user's stream to surface related content from that user's own stream(s).