What if recommendation algorithms were purely interpretable?

I used the OpenAI decisions API to generate a few hundred of questions about each movie/book/painting. A local model tries to predict your preference for each of those questions from your answers.