Converse with pdf with many tables

[twilit egret]

Hey #everyone,
I’m working on a use case that involves interacting with a large number of financial reports (PDFs). Even if we retrieve the right chunk of information, financial data often includes a lot of numbers in tables, and LLMs can sometimes hallucinate or misinterpret numerical values.

Has anyone explored alternative approaches to handling this? Are there any reliable methods to ensure numerical accuracy when extracting and reasoning over structured financial data? Would love to hear your insights!

Thanks in advance!

[ancient pendant]

What tools, tech are you using in this use case for data ?

[fast pagoda]

try Raptor and the Graph rag on the financial data

it can work

Additionaly , can you explain your method of parsing the pdf ? As afinancial report contains large number of table so how are you handling this ?

[grizzled cedar]

Following g