Zibaldone

The new release of tei-mcp introduces span-locked composition, a system designed to prevent the most damaging class of hallucination in AI-assisted TEI encoding: silent rewrites of the source text. The model never types body text; it registers tags as offsets over the source, and the composer refuses to return any TEI whose flat text content differs from the original by a single byte.

Most primary sources never get published. The grant-funded digital edition model is too expensive, too centralised, and too fragile to work at scale. Here’s why I’m building the critical edition of Cavriana’s letters differently.

tei-mcp is an open-source MCP server that gives AI coding assistants direct access to the TEI P5 specification — element lookup, attribute resolution, nesting validation, document validation, and ODD customisation.

An interactive demonstration of the ITA2 (Baudot-Murray) telegraph code that helps students grasp fundamental concepts of binary encoding and state machines

The Levante is an elusive toponym, as it is usually defined in relation with–or in opposition to–another territory. What was then Tuscany’s Levante in the 16th century? The data that I gathered from the MIA database gives an unexpected answer.

GPT-3 helps converting large amounts of bibliography into a database in a short amount of time

Pre-trained language models can help scholars automate some of the most tedious and labor-intensive tasks of edition. Based on the curated annotations of Secrets of Craft and Nature in Renaissance France, I evaluate the extent to which a model such as GPT-3 can be rapidly trained to annotate 16th-century technical manuscripts.

Newer version of the interactive treemap with additional features.

Interactive visualizations provide readers with an alternative sensorial input to navigate complex archival documents.

Dashboard web applications increase situational awareness in the archive, ultimately improving the latter’s accessibility and researchers’ productivity