I intend on building a TikTok clone, and the way it will replicate the TikTok algorithm is to have a lightweight image recognition model and a speech-to-text model run locally when you upload a video, and generate kind 1985 (?) labels. Then it runs a local algorithm that chooses notes based on your interests. You don’t need to know about all the notes that exist, only all the kind 22 notes with associated kind 1985 labels