Ask HN: Can RAG be used for recommendation system?
pinner Saturday, February 21, 2026I have already worked on this for a week, but it's very hard to define the needs. The idea was to create a fully customized feed (with privacy! nothing is sent to the server), across different platform. Here are some questions that I have encountered:
1. I'm using sentence-transformers/all-MiniLM-L6-v2 for speed (people will never wait seconds for a single page), and every "clusters" are scored based on historical likes/watch/dislike with my custom layout. The thing that I'm not sure here is whether people will accept my layout(for now I was transforming the HN into tiktok-like scrolling experience). 2. Without any data, it's very hard to calibrate the scoring system for the baseline. One thing that I am uncomfortable with current algo-based platforms is that they are rather ignorance on personnal taste. They recommend based on what "others" similars to my taste instead. Anothing is that we can not select the "exploring" level. I think only 10~20% of the feeds on X are "new" to the user, and this proportion is not adjustable at all by the users themselves. However, I am thus not able to get more exploring contents from these websites.
It's quite massy with these two questions, but I really don't know how to continue since there are no existing apps/tools for this at all. If you have any needs, please share it :)
Really need your interaction to see if I should continue on this project or not!!!