Reading by the shape of a story.

Most recommendation engines sort books the way a shop does - by genre, by author, by what sold well beside it. That tells you where a book sits on a shelf. It tells you very little about what reading it is like.

Readers do not work that way. Someone who loved enemies-to-lovers and found-family is not asking for any fantasy romance. They are asking for those specific shapes: the slow turn from hostility to devotion, the family assembled rather than inherited. BookForYou searches for those.

How it works

Describe a book in your own words and a fine-tuned BART model reads your description and writes out the tropes it contains - roughly seven of them, drawn from a curated vocabulary of 150. Sentence-BERT then retrieves the hundred closest books from a catalogue of 42,350, and those hundred are re-ranked by how much their tropes overlap with yours.

The overlap is deliberately one-sided. A book is never penalised for carrying tropes you did not ask for - only rewarded for carrying the ones you did. Readers want what they asked for satisfied, not exclusively present.

Because the tropes do the ranking, every recommendation arrives with its own reason. The matched tropes shown on each result are not an explanation generated afterwards to sound convincing; they are the thing that put the book in front of you.

What it does not do

It does not know who you are. There is no account, no reading history, no profile being assembled. Every search stands alone, which means a book with no readers yet competes on equal terms with one that has millions - and it means there is nothing about you to lose.

Built by Ritu Kanchi. The method is described in a paper published at RecSys ’26.