Card catalogs were how libraries tracked every book — one handwritten card per item, filed in drawers by author, title, or subject.
The BPL's rare books catalog (385 drawers, 453,006 cards) is still the most complete index of the department's holdings. New Vision Language Model based OCR makes them searchable for the first time — by semantic meaning (try “antislavery pamphlets”), not just exact words.
This is a finding aid: search retrieves the original index cards, which remain the source of truth. Browse a drawer to see what’s nearby — just as you would with the physical catalog.
Does VLM-based OCR actually improve retrieval over older Tesseract output? The tab runs the same query against both transcriptions side by side.