I'm not its target audience (not much of movies fan), but considering I made sheets for similar uses for music and games, it makes me think: couldn't it be fully offline? More reassurance of no data collection and it should keep being useful virtually forever, specially if physical media and DRM-free movies and series get traction. instead of an LLM, why not a more direct tag or like/dislike system, with either having a bit of randomness for suggestions to have an algorithm? in line with the first point, if an LLM is a must, why not have a local one trained specifically on movies and series?