AI helps add 10k more photos to OldNYC (danvk.org) AI

The developer of the OldNYC photo viewer says AI-assisted geocoding and OCR have helped add 10,000 more historic photos to the site, with more accurate placement and better transcriptions. The update uses OpenAI (GPT-4o) to extract locations from photo descriptions, relies on OpenStreetMap-based datasets instead of Google’s geocoding, and rebuilds OCR with GPT-4o-mini for higher text coverage and accuracy. The post also notes a migration to an open mapping stack to reduce running costs and allow historical map styling, while outlining plans to extract more image information and expand to other collections or cities.