langchain for search and summary of my pdf files.

[carrotcookie]

Hi, community

I have a question because there is a difficulty in creating a document-based chatbot as a school project.

---

I am saving a pdf file in chroma db in units of chunks using textspitter.

When a question is asked, it answers based on the vectorstore, but I want gpt to answer even if the vectorstore does not have information related to the question.

However, it is necessary to distinguish whether the answer was based on vectorstore.
This is because if gpt answered based on the vectorstore, I want to give the page information which part of the pdf document gpt referred to.

In other words, if gpt answers with reference to vectorstore, source_document will be given to the user, and if there is no data in vectorstore but gpt answered, source_document will not be given to the user.

And I want gpt to be able to remember the chat history and reply. chat history is stored in the db and always passes chat_history as a factor, but it seems that gpt does not remember the previous chat history well.

finally, It takes quite a while to get an answer, so I wonder if I can adjust the time.