davanstrien/bpl-card-catalog

BPL Card Catalog Search

Search and browse ~453,000 digitized catalog cards from the Boston Public Library's Rare Books & Manuscripts Department.

Uses AI-powered OCR (small vision-language models) to make handwritten and typewritten catalog cards searchable for the first time.

Features

• —Semantic search — find cards by meaning, not just keywords
• —Keyword search — full-text search across OCR transcriptions
• —Compare OCR — see old Tesseract vs new VLM OCR results side by side
• —Browse by drawer — navigate the physical organization of the catalog
• —Image lightbox — click any card image to view full-size

Stack

• —FastAPI + HTMX + Jinja2
• —LanceDB (vector + full-text search)
• —sentence-transformers (BAAI/bge-base-en-v1.5)
• —Dataset: Lance format on HF Hub