Agent Builder - Document Search for PDF, replying the reference page number of the answer.

[Wayne0119]

Hello, I am trying the Document Search app on agent builder, I upload five PDFs as source and ask question accordingly.

The agent can now answer my question(as the graph shows), but the "reference" it gives is a whole PDF,  which make it difficult for me to double check the answer(because my PDF size is big) .

Is there any way for the agent to also response information such as the page number of the reference? This could make double checking easier, thank you.

Best,

Wayne

[MarvinLlamas]

Hi @Wayne0119,

Welcome to Google Cloud Community!

It looks like you are encountering an issue with the Agent Builder's Document Search feature, which only provides the entire PDF document as a reference instead of specifying the page number where the answer is located.

Here are the potential ways that might help with your use case:

• File Chunking and Metadata: You may want to pre-process the documents before uploading them to Agent Builder. By splitting each PDF into smaller chunks and adding metadata (e.g., PDF name, page number(s)), you can then upload these chunks to Agent Builder and get accurate page numbers.

Regarding the use of Document Search app on agent builder to reply with the reference page number of the answer, I suggest filing this as a feature request. Please note that I can't provide any details or timelines at this moment. However, you may keep an eye on the release notes for any latest updates or new features related to Vertex AI agent builder.

You can refer to the following documentation on Google Cloud's Vertex AI: Parse and Chunk Documents. This guide explains how to pre-process documents by splitting them into smaller chunks and adding metadata for better information retrieval in PDFs:

• Parse and chunk documents

Was this helpful? If so, please accept this answer as “Solution”. If you need additional assistance, reply here within 2 business days and I’ll be happy to help.