Thanks for the elaborate response! I recently looked at doing something similar and ended up with the option of using vector embeddings. Is that something you've considered?
Thanks for the elaborate response! I recently looked at doing something similar and ended up with the option of using vector embeddings. Is that something you've considered?
For a first pass having some emb search would require we store and process a lot, which we dont do for this light weight demo. all the posts and recommendations are done on the fly, only looking at the current top page and re-ranking. as some others suggest, if we wanted to expand the recs more broadly than the front page then it might be a great way to run a first cut before running our rec llm on it.
For richer data to build a profile its something we look at a bunch for other projects, which could get folded in here if we decide to make it more persistent.