Comment by deforciant

have you tried other models to generate embeddings? I am going to that direction too to create an additional layer of helpers for search. Also, thinking if the document is not too big, it might fit into the initial context with the prompt