---
title: "Automatic reading of archive registers"
url: "https://edoardo.anteprime.edoardogabrielli.com/en/projects/archive-registers-reading"
lang: "en"
updated: "2026-10-31"
---

# Automatic reading of archive registers

> Automatic reading of archive registers and files with a local vision model, human review and page citations.

## Automatic reading of archive registers

A local vision model reads registers and files, and a person reviews every piece of data.

## The problem

Archives hold paper registers and files full of useful information, but as long as they stay images nobody can search them. Retyping them by hand is slow.

The goal was searchable data without giving up reliability: every piece of data must be traceable to the page it comes from.

![Handwritten letter from 1849 (stock image)](https://edoardo.anteprime.edoardogabrielli.com/media/originals/2026/10/a8525779ff3f4968af167691c65a77d5/lettera-manoscritta.jpg)

## What I built

### 1. Page-by-page reading

A vision model running locally reads each page and extracts the fields: 18 for the registers.

### 2. Human review

A person checks each piece of data next to the page image and corrects it if needed.

### 3. Page citation

Every item carries a reference to its source page, so whoever uses it can verify it.

### 4. Working interface

A web application to upload files, follow progress and review results.

### My role

Developer at ETT. I worked on the reading pipeline, the service and the review interface, in collaboration with Scripta, ETT's digitisation line. The institutions involved are not named.

### Technologies

Python · FastAPI · vLLM · vision model · React · Next.js

> The model makes different mistakes from a person, and the review has to be designed around them: showing the data next to the page costs little and saves a lot. Running the model locally taught me to deal with memory and speed, not just accuracy.

— What I learned

[Scripta on the ETT website](https://ettsolutions.com/scripta/)

## Registers to make searchable

Tell me what kind of documents you have: we start from a sample and see how well they read.

[Get in touch](https://edoardo.anteprime.edoardogabrielli.com/en/contact)

[← All projects](https://edoardo.anteprime.edoardogabrielli.com/en/projects)

Stock image: letter by Frances Appleton Longfellow, 1849, public domain (Wikimedia Commons). It does not show the project.