Greenlight

Inspiration

I upload a lot of videos for content creators in the true crime and ASMR space and we are currently having a problem with YOutube monetisation

What it does

Drop a finished cut in. It transcribes locally, finds the moments that cost you ad money, cites the published rule each one trips, and writes a cut list you can run.

How we built it

Next.js and React, Whisper in the browser on WebGPU. Each platform is a YAML pack the engine reads. The engine is DOM-free and network-free, so all of it is testable.

Challenges we ran into

The failure mode was silence, not errors. Four bugs all looked like nothing happening. Stacked mitigations also buried a graphic passage from severity 4 to 2, so I capped them at one.

Accomplishments that we're proud of

The generated cut list actually runs. The flagged window went from -20.5 dB to -91.0 dB and later speech survived at -20.3. Same clip reads limited on Instagram, cleared on YouTube.

What we learned

A tool that reads policy goes stale quietly. It now checks the source pages daily and says 1 of 3 sources, not all, because two platforms block reads. Honest beats impressive.

What's next for Greenlight

Get Instagram and TikTok into the daily policy check. More platforms as packs. Read what is on screen, not only what is said. Then clear a whole back catalogue in one go.

Built With

• nextjs
• react-native
• whisper
• yaml