DocMaster Keeps Document Hierarchy Instead of Flattening PDFs Into Chunks

arXiv·medium signal

DocMaster is a hierarchical structure-aware system for LLM document analysis over academic papers, technical manuals, and financial reports, where users must first filter relevant documents from large collections and then do deep QA over the subset. Its core claim is that existing systems flatten documents into plain-text chunks and discard rich hierarchical structure (sections and their nesting), which DocMaster preserves. Relevant to anyone building RAG or document-intelligence pipelines that currently chunk blindly.