I think it's possible. One problem that I have with Remembrancer, as well as all approaches based on the same basic idea of 'show the user a list of similar ite

I think it's possible. One problem that I have with Remembrancer, as well as all approaches based on the same basic idea of 'show the user a list of similar items', is that this is fundamentally guaranteed to be mostly useless because the user will already remember many similar items and have deliberately omitted them. Sometimes you will forget, and then it can be quite helpful, sure, but for the most part, this retrieval strategy is inherently limited. What you really get serendipity is from re-encountering things you don't know or have forgotten. And we know how to predict if you have forgotten something: spaced repetition and the forgetting curve! To manufacture serendipity, you simply track 'flashcards' of important things, but optimize for the opposite of the normal spaced repetition goal: you want to review only things you are predicted to have forgotten by now. https://gwern.net/note/statistic#program-for-non-spaced-repetition-review-of-past-written-materials-for-serendipity-rediscovery-archive-revisiter https://gwern.net/note/statistic#program-for-non-spaced-repe... Then, as a Remembrancer agent, you make it show preferentially items that are similar but you have probably forgotten. (The Remembrancer agent can be treated as a fraction of a review, on the grounds that even just showing you the title briefly can trigger serendipity and if it didn't provoke any reaction from you, then that is evidence it is not useful and that title should be downweighted and show up less.)