Artificial Intelligence
Background
For many years, we have been developing publishing workflows with a strong focus on efficiency and quality. We use AI in production wherever it can reliably accelerate or improve individual steps. At all points where review, judgment, and responsibility are required, we rely on our colleagues. This page presents three current AI-supported applications from our ongoing production work, as well as our consulting services for publishers.
1. Human-in-the-loop
In production workflows, AI handles clearly defined tasks. Anything that cannot be reliably reviewed or adjusted by AI is flagged and passed on for manual review. This is not a safety net added at the end, but a deliberate part of the design: fast where reliability allows it, careful where care matters.
2. Quality first
We compare new workflows with our established processes and manually reviewed reference data. If quality or consistent reproducibility does not meet our standards, we do not switch our production to an AI-supported process at that point, except at the explicit request of a customer.
3. Data in the EU
Our AI workflows are GDPR-compliant and use EU-in-zone routing. Manuscripts, data, and model requests do not leave the European legal jurisdiction. They are developed and operated in Leipzig by the same experts who have been shaping le-tex and the publishing services industry in Germany for 27 years.
Consulting
AI raises three important questions for publishers that we can help answer: What is actually worth doing? What can already be operated reliably today? Which risks – data protection, rights, hallucinations, and traceability – need to be clarified in advance? We bring the experience we have gained from our own production environments and are transparent about where the limits lie. Typical topics include the selection of models and providers, on-premises or API-based deployment, evaluation and quality metrics, and integration into existing editorial, XML, and typesetting workflows. The right approach depends on the use case. Not every task requires a large LLM. For clearly defined tasks, smaller specialized models, traditional NLP, or rule-based methods are often more reliable and more cost-effective.
We configure workflows according to the requirements of the respective publisher, taking into account data protection, rights issues, and the requirements of the GDPR and the EU AI Act.
»The potential for AI-driven automation in publishing services is enormous – and we are using it. But the benchmark remains the same: the highest possible quality, consistent reproducibility, and uncompromised scholarly integrity.«