My MA thesis, Loud Yet Invisible: A Humanist-Designed Pipeline for Unlocking the Early Modern Archive, built a pipeline for reading a corpus of about 20,000 printed documents (roughly two million pages) from 1614 to 1810, with ginger as a case study.
The pipeline runs olmOCR transcription, then EarlyModernNER, a custom named-entity recognition model, then a knowledge graph, then GraphRAG retrieval. It is designed as a discovery, search and ranking tool for historians rather than an automated answer machine, and it includes methodological work on transcription error rates and gold-standard ground truth for early modern documents.