A film recommender that’s only yours.
palette.film is a personal-taste film recommender. You hand it your Letterboxd history; an ensemble model learns what you actually like and scores every film you haven’t seen by what you’d probably rate it.
Other apps recommend what people like you tend to like. palette predicts what you would actually rate it. Different problem, cleaner answer.
How it works
Upload your Letterboxd export. We hydrate every film with metadata from TMDB — genres, cast, crew, keywords, similar titles — and train a model on the films you’ve already rated. Ten independent components vote on each unseen film, and the weights shift to favor whichever components actually predict your stars best. Your top weight may not be anyone else’s.
Every prediction is interpretable: the model can show you which components produced a given score. It’s also small, transparent, and re-trainable — your taste model is a couple of megabytes of math, not a billion-parameter black box.
Credits
palette.film is built on top of The Movie Database. Every film poster, cast list, keyword, and similarity edge in this app comes from TMDB. We’re grateful for the open API and the volunteer community that maintains the catalog.
Built by Jack Bolden. Reach out at jackbolden23@gmail.com.