A manufacturing company builds a RAG system over equipment maintenance manuals. Documents are 200-page PDFs with dense technical illustrations, schematics, and parts diagrams. Text extraction from PDFs produces incomplete content because critical maintenance steps are embedded in illustrations (e.g., 'Turn valve A clockwise 3 times' is written inside a schematic diagram, not in the surrounding text). Text-only RAG misses these steps. What chunking and indexing strategy captures this content?

Multimodal indexing via vision-based description generation is the correct solution for PDFs where critical information lives in images and diagrams. Rendering pages as images and using Claude vision to generate comprehensive text descriptions captures valve labels, step numbers inside diagrams, parts references written on schematics, and other illustration-embedded text that standard text extraction misses entirely. These descriptions become the indexed content, making illustration-embedded information retrievable through normal semantic search. Option A (manual transcription) is operationally prohibitive for a 200-page maintenance manual corpus and doesn't scale. Option C (page-level indexing) keeps illustrations and text co-located but doesn't extract the illustration content — retrieval can surface the page, but the illustration content is still not parsed. Option D (filtering illustration pages) directly causes the described problem — maintenance steps on illustration pages are completely lost from the index.