From Page Images to TEI: Schema-Guided LLM Transcription of Complex Literary Documents

Slides presented at the seminar From Thought to Practice: A Dialogue Between Artificial Intelligence and Digital Humanities (University of Bologna, 5 November 2026). The presentation introduces a schema-guided pipeline for transcribing historical printed and handwritten documents and converting them into TEI XML using multimodal large language models (LLMs). Document-specific Pydantic schemas guide structural recognition and produce validated JSON; Python processing and configurable YAML rules assemble page-level results into TEI. Examples include Italian opera libretti, poetry collections, annotated editions, and correspondence, with particular attention to cross-page structures, reusable document models, and scalable digital corpus creation.