#VDocRAG

'Vision-RAG outperforms Text-RAG on visually rich documents by preserving layout and figure grounding, offering measurable end-to-end gains while requiring selective fidelity to control token costs.'