Attenbot | Devpost

Inspiration

YouTube is not incentivized to keep your kids from brain rot like Mr. Beast and Ryan's World. This is how YouTube makes their money.

The only alternative to protect your kids is to pre-approve every video manually (a feature that YouTube hides anyway), which is extremely tedious and not feasible for most parents.

We want a world that's not skibidi toilet and is aligned with nurturing the next generation, so we decided that this problem would be the focus of our hack for impact.

What it does

It uses an LLM to filter the videos a child sees on YouTube dynamically based on the parent's filters/values. The parent defines things like "I don't want anything that's clickbait whose purpose is to take my child's attention" and "Joey has been interested in dinosaurs in school so emphasize videos like that this week", and Attenbot takes care of the rest. Your child will have a clean feed whenever they open up YouTube.

The parent gets emailed weekly updates on the videos their child is watching. This provides an opportunity for the parent to refine Attenbot's choices, helping it better decide the content that meets the child's needs. Parents can now avoid relying on the YouTube algorithm, which does not prioritize the well-being of their child.

How we built it

We have a Chrome extension that dynamically blocks the content on YouTube and makes API calls to our NextJS server endpoints to get the filtered videos. The filter is based on the structured output from gpt-4o given the YouTube videos that the child would have seen without the filter. The filter prompt is seeded with the filters defined by the parent and by the negative/positive examples that the parent defines from their weekly reports.