· DHInfra.at · Use Cases · 2 min read
Information Extraction from Monastery Account Books
Testing the DHInfra.at inference API on the account books of Aldersbach Abbey — open models, matched against commercial ones.
How well can a language model automatically classify and extract information from early-modern monastic account books? A team working on the Rechnungsbücher of Aldersbach Abbey put this to the test on the DHInfra.at infrastructure.
The pipeline is straightforward: prepare the data, send it to the inference API, receive structured output, and feed the results back into research and edition work. The GPU cluster processes the requests on demand.
Direct API access as an easy substitute
During the development phase we deliberately kept access simple: researchers reached the cluster’s OpenAI-compatible inference API directly (an SSH port-forward standing in for the eventual direct HTTPS endpoint). A single key is enough to identify the project — everything else runs remotely. No cluster login, no local GPU, no heavyweight setup.
The results were encouraging: outputs from open, state-of-the-art models were on par with or better than commercial offerings, and a single set of experiments saved several hundred euros compared to paid APIs. The main node comfortably produces several thousand tokens per second.
What comes next
We are turning this into a reusable playbook: documentation on our GitHub on how to run this kind of API-based information extraction yourself, and a ready-made standard notebook on the cluster so new projects can start from a working example rather than from scratch.
This use case was presented at DHd2026 (Vogeltanz, Vogeler, Klugseder, Rufin, Unterholzner, Spoerer, Pößniker, Kern, Kröll & Lübbers: Kloster trifft KI, DOI 10.5281/zenodo.18702943).
