Skip to main content
leapedge.clip

Pipeline walkthrough

How leapedge.clip works.

Every analysis runs the same six-stage pipeline. Fast-model triage, smarter-model synthesis, and a self-critique pass before anything reaches your dashboard.

  1. 01
    stage

    Paste — any YouTube link

    Drop a watch link, share link, or shorts URL. We read the title and channel from YouTube's oEmbed API — nothing to configure on your end.

  2. 02
    stage

    Transcribe — captions first, audio fallback

    Three caption sources are tried in order. If all three come up empty — creator-blocked, language mismatch, or no captions at all — our AI watches the video itself on Google's servers and writes down what it heard. No local download, no yt-dlp, no special setup.

  3. 03
    stage

    Extract key points — fast model triage

    A lightweight Gemini Flash-Lite model reads through the transcript and pulls out roughly ten key points from the video. Quick and cheap — its job is to give the next stage something focused to work with.

  4. 04
    stage

    Draft insights — smarter model synthesis

    A more capable reasoning model drafts the trading thesis: tickers, direction, conviction level, and a verbatim supporting quote for every claim. Every insight has to point to something the creator actually said — no paraphrasing.

  5. 05
    stage

    Double-check — self-critique pass

    A second pass audits each insight against its quote: does the quote actually support the claim, or did the model drift? Anything that doesn't hold up gets removed before your dashboard sees it.

  6. 06
    stage

    Verify — every quote links to its timestamp

    After the double-checking pass, each quote is matched against transcript segments to find the exact moment in the video. Click any quote in the analysis view and you land right at that second.

Ready to try it on a video you've been meaning to watch?

Start research