Chapter markers for recorded lectures

DesignEasy30 min

Your task

Design a feature for an online course platform that splits each recorded lecture into chapters, each with a start time and a short title, using a hosted LLM on the lecture's transcript. The transcript arrives as timestamped segments a few minutes after a video finishes processing. Instructors upload about 1,500 lectures a day; most run 45 to 90 minutes, and the longest workshops run three hours and do not fit in the model's context window. Chapters should appear within 30 minutes of the transcript being ready, instructors can edit them, and every start time must point at a real moment in the video. Describe the processing flow, how long transcripts are handled, and how you keep the timestamps honest.

It covers

  • Chunking
  • Grounded output
  • Job retries