With the embedding and Pinecone integration on display, the last piece there is to show someone how to do content recommendations. Fortunately, Pinecone has this in fairly good detail:
The way it is operating currently:
This module itself cannot provide an OOTB content recommender experience because that relies on specific implementation around user data, which may come from anywhere. For example...
If you wanted to leverage this and create a content recommendation service, you would need to do something like:
This pattern is already somewhat in place as a demo right now with the "similar title checker" in the openai_content module. It takes the current node title in the form and vectorizes it, then searches Pinecone. Content recommendation is similar, but needs user fed data. This would also be similar to content personalization too.
I can see this benefitting people who don't have Solr/ES/Algolia/Personify+++ or similar at a far lower TCO. As long as it returns 'good enough' results, it seems compelling to me. But we just need to document an example of 'how' you'd do this.
Way down the line, I can see the potential for plugins with configuration entities so admins can define different recommendation criteria (what user data to draw from) and then apply those to a Block instance for example - or possibly integrate to Views and use that same UI (somehow) to do the same.
Postponed
1.0
OpenAI Embeddings
This might be better suited for Search API AI.
I have initiated this module https://www.drupal.org/project/recommend_ai → for the same. Although it's in very early stage but the goal is to have a full fledged content recommendation system with personalization