An advanced RAG pipeline that improves multimodal responses by creating context-aware image summaries and using the generated text response to guide image selection, overcoming common context loss issues.