Researchers conduct first comprehensive study on how PDF parsing and chunking strategies impact RAG system performance for financial document question answering.

Study addresses the challenge of automated PDF processing, which is difficult due to heterogeneous content including text, tables, and images designed for human reading

Research focuses on Question Answering as a specific language understanding task using two financial domain benchmarks, including TableQuest—a newly created publicly available benchmark

Systematic evaluation examines multiple PDF parsers and various chunking strategies with different overlap configurations to identify optimal design choices for RAG systems