Thanks for asking. Transcript inputs are from YouTube but the actual segmentation part is done by a machine learning model. It analyzes the context of a given transcript to separate the transcript into different segments, and then comes up with topics that represent each segment.