I am reading a single billing PDF and then creating embeddings and storing them into vector stores and then calling Lang chain with prompts using LLM to generate the responses. This works fine for single PDF. When I upload multiple PDFs and then ask questions, the solution fails as it doesn’t recognize which page in the multiple document it must refer to and the context information is lost. how to fill this gap?
Please share any input to store contextual embeddings while reading the text chunks and text splitter and store them into vector store. Please advise and share some samples.