Best approach for extracting data from scanned PDFs with mixed text, tables, and images?

Our documents are scanned PDFs — some text-layer, some image-only — with embedded tables, signature blocks, and handwritten annotations. OCR alone misses structure. Vision models are expensive and slow. What's the practical pipeline for high-throughput extraction that preserves table relationships and doesn't hallucinate values that look plausible but aren't in the source?