Design a video content analysis platform

Creators upload videos and ask for them to be analyzed: scenes, objects, a speech transcript, safety labels and embeddings, all tied to timestamps in the video. Design the platform that does this asynchronously.

Cover:

• uploading large files so an interrupted upload can resume, and starting each analysis only once
• running the analysis stages and publishing a complete set of results, where every result records the pipeline and model versions that made it and a retry never leaves duplicate or mixed-version output
• showing job progress, fetching the results for a time range, and searching across a user's own analyzed videos