Millions of natural history records lack spatial metadata yet contain text locality information. We present an open-source pipeline for mass georeferencing at scale, combining deterministic spatial methods with LLM-based approaches for parsing locality strings and characterizing spatial relationships.