Three and a half years of DHInfra.at: results, and the handover to regular operations

Today, at our closing event, we presented the results of three and a half years of work. This post is a written summary of what we showed — a snapshot as DHInfra.at reaches the end of its funding this month and the work moves from a project into regular operations.

DHInfra.at by the numbers

DHInfra.at ran from 2023 to 2026 as an infrastructure project — not a research project — under the funding programme “(Digital) Research Infrastructures”, supported by the BMFWF and the ERC Recovery and Resilience Facility (NextGenerationEU). In that time it brought together:

• 9 institutions across Austria, coordinated by the University of Graz
• 4 working groups and 8 work packages, against 28 milestones
• 50+ members and 100+ associates

The goal was to close gaps that individual institutions could not close alone: standard digitization in libraries, archives and museums; generic repositories and databases; and access to high-performance computing outside the natural, technical and life sciences. DHInfra.at addressed this in four areas.

What was built

1. Advanced digitization and data capture

Two portable multispectral imaging (MSI) systems — XpeCAM and MISHA — were acquired and are made available to partner projects through a loan-and-training model, backed by GPU hardware for processing (see the digitization resources). They have already recovered text from heavily damaged material, including the medieval charters of Cetinje, Montenegro. On the high-throughput end, a robotic scanning line reached production use, capturing 400,000+ historical registration slips at an error rate below 1%.

2. Databases, repositories, IaaS and SaaS

The University of Salzburg’s Database Research Cluster (Proxmox + Ceph) now provides DH projects with virtual machines, database backends, and reproducible development and test environments via an integrated toolchain (Ansible provisioning, CI/Git runners) — see servers, services and storage. Test cases ranged from the long-running MHDBDB medieval-German database to the Pangloss prosopographical prototype, validating the cluster for real DH workloads under load.

3. Domain-affine machine learning and compute

GPU clusters in Graz and Krems were procured, installed and brought online — state-of-the-art NVIDIA nodes (H200, RTX Pro 6000, L40S) on an InfiniBand interconnect, reachable across the academic network; the Graz cluster install is documented in its own post. Beyond classic HPC elements (single sign-on, SLURM batch scheduling, interactive JupyterHub), the distinguishing feature is a cloud layer: containerization plus hosted, OpenAI-compatible LLM and embedding APIs backed by large open-weight models — so projects can call AI services without their data leaving the academic network.

The use cases are concrete: LLM-based information extraction from medieval accounts and from the Managing Maximilian corpus, LLM fine-tuning and inference for historical research, OCR and segmentation for newspapers (the ÖNB’s ANNO corpus) and historical job ads, cadastral-map segmentation, the Unlocking the Schematismus pipeline, and experimental from-scratch model training such as the HabsburgLLM teaser.

4. Free and open-source software

Three DH tools received sustained investment — clearer documentation, easier setup, and active development (overview on the open-source software page):

• Pletka — collaborative ontology editing with a unified format and export to RDFS, SHACL and more
• QLever — a high-performance SPARQL engine, with a complete documentation rewrite, unified use-case Qleverfiles, and official packages for Debian/Ubuntu/macOS
• liiive.now — real-time collaborative IIIF annotation, self-hostable and MIT-licensed

Where things stand now

Right now, almost all of the planned infrastructure building blocks are already heavily in use. The GPU cluster and the LLM/embedding APIs are on the verge of rollout to CLARIAH-AT partners, and the governance core — operating concepts, a sustainable business model, and a helpdesk approach — is in place. The dhinfra.at site itself is being relaunched as the internal test phase ends.

What comes next

The DHInfra.at project closes at the end of June 2026. From July, a CLARIAH-AT–funded follow-up — DHInfra-Gov — takes the work from “project” to “regular operations,” focused on governance, rollout and live deployment of the infrastructure to partners. Further milestones on the horizon include a clariah.at relaunch later in 2026, national sustainability via CLARIAH-AT and dedicated service desks, and an international route via HORIZON.

There is no public platform link to share just yet — stay tuned. In the meantime, questions are welcome at dhinfra@uni-graz.at.

With thanks to the IT services of all partner institutions, the wider ASHPC and DHd communities, and the BMFWF and ERC-RRF for making the project possible.