Building Document Structure with Loop Engineering: Recovering a PDF’s Outline from Body Typography for RAG

Many documents lack a structured table of contents, which is a critical component for high-quality Retrieval-Augmented Generation (RAG) systems. A novel technique reconstructs this missing outline by analyzing typographic cues directly from the document's body text. This method identifies potential headings using six deterministic signals, including font size, boldness, numeric prefixes, and line spacing. An LLM then validates these candidates, filtering out false positives to produce a clean and accurate document structure. The final result is a standard table of contents that enables more precise, section-aware retrieval and chunking within the RAG pipeline.