I am working on a streamlit app at the moment where I want to have the user upload a (larger) 300 page ~20MB document. the idea is that the document can be reviewed against the training knowledge (retrieval) and preset instructions/possible prompts. I have it up and running, the assist works fairly well and besides that, I am struggling with the document uploader. do I need to embed it? or utilize a vector database? not sure with the assistants. anyway looking for any insights or pointers here. thank you!!