Donald Murre

I am a tax lawyer and an independent researcher. I study where AI systems give confident answers they should not, starting with retrieval for law and tax. I have written Python for ten years, including NLP, machine learning and trading algorithms. I hike in the mountains.

LinkedIn · Google Scholar · ORCID · Contact

Currently working on

• Retrieval for law and tax that flags when a source does not apply (working paper)
• Privacy-preserving use of cloud LLMs on confidential legal documents: in-browser pseudonymisation, own multilingual NER models, homomorphically encrypted checks, egress-level evaluation
• Land-cover detection from aerial imagery in Switzerland, as an audited service with structured, calibrated, evidence-carrying output
• Knowledge graphs and automated pipelines that build sourced content
• Species-accurate image generation and verification for more than 230,000 European animal and plant taxa: LoRA and textual-inversion fine-tuning of diffusion models, BioCLIP and vision-language jurors
• Fine-tuning and evaluation of encoders (multilingual NER, text-origin classifiers) and local LLMs
• A self-supervised financial model on trade and quote data, against standard volatility models
• Interpretability experiments on a small open language model

More on the research page

Publications

A working paper on reliable retrieval, and peer-reviewed articles and case notes on tax (2009–2013), all with DOIs. See the full list.