I think if you allow a set of YouTube videos as input, it'll be quite powerful coupled with transcription ability of LLMs. Lots of people consume content that way. As an added bonus, you can show the performance summary about the sections the user did well or not so well on with video links to those timestamps for them to go back and review.