Multimodal RAG & Pipelines

Build RAG over real documents with images and layouts — why text RAG breaks, three multimodal architectures, the modality gap, page-as-image retrieval with ColPali, grounding a VLM, and evaluation.