quoting source and page number

[pcwerk]

Is it possible to embed document filename and page number with docling?

[dolfim-ibm]

What is your definition of "embed" here? Do you have some concrete example?

[pcwerk]

Ah, yes. Apologies for the terseness of the inquiry. Maybe I can best describe it as a use case for using Docling (which, by the way, so far appears to be one of the best PDF extractors out there based on my unscientific blind mouse evaluation).

We have 5 PDF documents. I would like to load these documents into our vector store (Chroma, in our case) database using Docling. Docling does the extraction of text and tables really well, and so I want to leverage this when embedding it into the vector store.

The goal is that I'd I like to associate filename and page number into the vector store so that when the retriever finds matching text in Chroma, I can cite the source document and page number.

[dolfim-ibm]

Got it.

We are actually covering this in the llamaindex and langchain extensions directly. See the chunking section in the README and and the examples notebooks.

[pcwerk]

Thanks @dolfim-ibm appears to be what I was thinking.

[GOVINDPURACHERY]

Is this chunking option available in the "RAG with Docling and LangChain" pipeline???? Currently, it shows that "RecursiveCharacterTextSplitter" from "langchain_text_splitters" is being used for chunking; however, the results do not include page numbers or bounding box information. Can I utilize the aforementioned chunking method (HierarchicalChunker) in the LangChain-based RAG pipeline?

[Sivaprasad04]

@GOVINDPURACHERY Yes this would be very helpful :)

[vagenas]

Yes, you can see an example in this notebook.