The paper presents D-RAC, a technique that normalizes heterogeneous enterprise documents into PDF before converting them into retrieval-optimized Markdown using multimodal LLMs, preserving structure while avoiding costly agentic chunking. It reports a 95.7% reduction in output tokens and 77.8% cost reduction versus frontier-LLM-based chunking. The method was evaluated on 236 documents (795 pages), processing the full corpus in 72 minutes with zero errors.
