typesense

Hi,

I have a database where I store user sessions, eg I have the documents that I'd like to generate recommendations from.

Here's an example structure from my DB:

User123: {
   $id: 0123ADWWA
   name: David,
   userSessions: [ {
        $id: oijOIJWDOA
        sessionDate: "2024-04-10"
        userSessionActions: [
                              {
                       $id: alsdmfaw
                       actionType: "view_product"
                       productId: nmoiadoijad
                 },
                  {
                       $id: 12323pkak
                       actionType: "view_product"
                       productId: alkdADW97
                 }
         ]
 } ]
}

My impression was that I could plug in all these ids from the aforementioned documents from my user sessions into a vector search and get "recommended" documents from typesense. My typesense collection has auto embeddings that are generated with OpenAI using the text-embedding-3-small model with 256 num_dim.

I have already seen the article you guys made on recommendations and how we can use starspace or transformer4rec? I'm unsure why I can't do a vector search based on my current ids I gather myself?

Am I mistaken?