Klimabrølet

Klimabrølet ran a site where people uploaded their own roar for the climate to a public gallery. Anyone could log in with Vipps and upload, but nothing appeared until an admin approved it.

I built large parts of the admin solution volunteers used to make sure the roars were not political, hateful or otherwise inappropriate, and extended the site so a video could be viewed as an image.

The moderation queue was the real problem. I set up the Azure Image Recognition API to automatically filter inappropriate videos before a human ever saw them, then wrote a feature versus price report on the approach, which cut the cost of using image recognition by 60 percent.