cat README
Research
I work at the cutting edge of applying AI, not of building it. Frontier models are built by people far smarter than me. I work on what comes after: making these models useful, checkable and honest in real work.
Focus
Extracting data and insights from large amounts of unstructured text: patents, reports, policy documents, interviews. Less "ask the chatbot", more "read a hundred thousand documents carefully and show me the evidence".
Methods: large language models, information extraction, semantic embeddings, community detection and hierarchical knowledge graphs, in human-supervised and agentic workflows that trace every result back to its source.
Current work
- PhD, Mapping Cross-Domain Solution Spaces: a hierarchical graph of generalisable solution concepts derived from public patent documents, linked to solution concepts from biology. In short: has nature already solved this problem?
- Framing detection: how well do fine-tuned BERT models and few-shot prompted language models detect threat and solution framing in German climate news? Measured against a human-coded reference set, with known error rates instead of impressions.
- Knowledge discovery: interdisciplinary, large-scale data analysis at the Chair of Systems Science under Prof. Ulrich Frey.
Publications and talks
- 2026 · preprint Comparing BERT Sentence-Pair Classification and Few-Shot LLM Prompting for Detecting Threat and Solution Framing in German Climate News Adam, R., Maier, D. & Kogler, M. arXiv:2606.26489.
- · poster Automated Detection of Threat- and Solution-Oriented Climate Change Framing in German-Language Newspapers 26th Austrian Climate Day, University of Vienna. Programme
- 2025 · master's thesis Automated Detection of Threat- and Solution-Oriented Climate Change Framing in German-Language Newspapers Maier, D. University of Graz.
Why this matters for companies
The methods are the ones companies need: turning their own documents into structured knowledge, and knowing how far to trust the result. Questions about data.