The page and the repo contain more context (give it a read, even if just for the buttery puns!). But, in short: It's quite crude. I extract captions from YT vi

The page and the repo contain more context (give it a read, even if just for the buttery puns!). But, in short: It's quite crude. I extract captions from YT videos using a private YT API (web scraping was a huge pain ;/), then I massage them a little bit and use GPT to process the timestamps. A waaaay better approach is possible, but I'll need to see if it's worth pursuing first. PS. SponsorBlock has a huge timestamp DB so I'm considering reaching out to the authors to see if we can create an agent using their extension and aiding their crowdsourcing effort.