News ยท Document AI

58,000 pages described in a local large-scale run

For BHIC, 2dA completed a first large-scale run on 58,000 pages of death certificates. Multiple local models worked together in targeted steps, entirely offline and on our own infrastructure.

View the original LinkedIn post

2dA AI Governance Platform and large-scale archive description run
Document AI

58,000 pages processed entirely locally

The output was built under control and delivered in a format that can be imported directly into Memorix without intermediate steps.

A large-scale run grounded in archive practice

An update from our development department: what we are learning from our first large-scale run...

At 2dA, we completed our first large-scale run for the automatic description of 58,000 pages of death certificates commissioned by the Brabants Historisch Informatie Centrum. An important part of this assignment was also the collaboration with BHIC, which gave room to build this route carefully and in real practice.

Not one model, but a targeted model route

In this production line, a system first reads the certificates. Then for each step it is determined which model best fits which specific action. So we do not let one model do everything; instead, we deploy multiple local models in a targeted way on the parts where they are strongest.

Non-public material remains offline

Because the material was not public, the full run was carried out completely offline: local, on our own infrastructure and without any connection to the internet.

Quality and direct connection to Memorix

The description reached a high level of quality. The output was also delivered directly in a format that can be imported into Memorix without intermediate steps.

Every run improves the system

What we learned again in this project is immediately taken into the further development of our system for automatic mass description. That is exactly the core for us: every run delivers not only output, but also new insights that help us improve the system further.

In this way, we are building step by step towards a system that can not only describe at scale, but can also do so carefully, locally and under control.

Next step

Would you like large document collections described locally and under control?