A Robust Two-Stage Retrieval-Augmented Vision-Language Framework for Knowledge-Intensive Multimodal Reasoning and Alignment. CDIS 2026, 2 (2), 42-52.