Search and folder contextualized content recommendations

[rtibbles]

Background

We've worked to develop deep learning models for providing content recommendations. These models use contextual information, such as the title, description, and metadata of a topic/folder content node and its ancestors, to provide content recommendations from our public catalog of channels. Eventually, the end goal is to use these capabilities for curriculum alignment by taking a curriculum skeleton (parsed from curriculum documents) and filling in the structure with content using this recommendation engine.

Story

As a user opted-in to new 'AI' Studio capabilities, who is interested in curating the most applicable resources from Studio's public catalog for my channel, I would like the ability to obtain a list of recommendations for a channel folder I'm editing. I would like the ability to then import selected resources from those recommendations which I find applicable to my channel and its folder.

Requirements

• All UI features should be controlled by the 'Al' capabilities feature flag
• Deep learning models should be hosted on HuggingFace
• The backend should properly restrict outbound API calls to HuggingFace by validating the user has the feature flag enabled

Out of Scope

• Building the feedback architecture to track implicit and explicit feedback on recommendations, except for integrating that architecture into search recommendations

[vkWeb]

Looks promising, I'll be happy to help with this! ❤️