[Python] How to send metadata to the LLM together with docs and context?

How can I include a document's metadata from vector db (Qdrant in my case), and send them together with context to the LLM (OpenAI in my case). For example, I store anonymized resumes in my vector db and each one has metadata with source (name of the file), how can I include it, and ask OpenAI to answer me with this metadata? Example payload (small part of page content preserved):

{
"metadata": {
"page": 0,
"source": "C:\\Users\\username\\Desktop\\resumes_dataset\\50328713.pdf"
},
"page_content": "Predicted Performance of Multilayered \nMetal-Mesh Screens. SPE Drilling & Completion. SPE-178955-PA. https://blablabla.org/10.2118/178955-PA."
}


Example call to the LLM:

def run_llm(query: str):
chat = ChatOpenAI(verbose=True, temperature=0)
qa = RetrievalQA.from_chain_type(llm=chat, chain_type="stuff", retriever=qdrant.as_retriever(),
return_source_documents=True)
return qa({"query": query})


print(run_llm(query="Which unique resume has the longest working experience as accountant?"))


It does a great job finding the correct answer, but I can't make it answer me with the document name, only by including return_source_documents=True does it show me where it found an answer.

Continue reading...