Methods, systems, and non-transitory computer readable storage media are disclosed for generating meeting insights based on media data and device input data. In one or more embodiments, the system analyzes media data and inputs to client devices associated with a meeting to determine a portion of the meeting that is relevant for a user. In one or more embodiments, the system generates a meeting summary, meeting highlights, or action items related to the media data to provide to the client device. In one or more embodiments, the system also uses the summary, highlights, or action items to train a machine-learning model for use with future meetings.