Transkribus for Census: Case Study of the 1897 Russian Imperial Enumeration

The 1897 Russian Imperial Census holds vital records for the study of Jewish life in the late Tsarist Empire — but its tabular structure, fading ink, and script complexity make it a formidable HTR challenge. This talk documents the full development of the PI Russian model from 32.92% to 9.75% CER, including the use of artificially generated training sheets to overcome ground truth scarcity, opening the door to mass analysis of this extraordinary historical source.