Create an automated pipeline that extracts structured knowledge from mixed-media documents (PDFs + images) into a knowledge graph. Current RAG solutions often struggle with interpreting visual data alongside text.