Multi-page PDF document extraction (Phase 2)

FeatureSecurityService
Shipped
July 14, 2026 at 9:12 PM UTC
Author
Kamo
Commit
b76ba7d

The vision path was image-URL only, so a borrower's multi-page PDF (bank statement, tax return) couldn't be read. Now: - VaultDocPublicController serves ?page=N by rendering that PDF page to PNG on the fly (via ConversionService /render-page) — provider-agnostic, since many vision providers accept only an image URL, not a raw PDF. Non-PDF/no-page requests are unchanged. Img→MinIO load factored into a shared VaultDocLoader. - DocVisionExtractionService reads a PDF page-by-page (capped at 15), merging fields across pages by keeping the highest-confidence reading of each path. A single image is still read as one page (unchanged behavior). - ConversionPageClient calls ConversionService (reuses conversion.service.url).

All changes

Like what you see shipping?

Every one of these updates lands in your workspace automatically. Start free and watch it grow week after week.

Start Free ForeverView Pricing