Donald Murre

I work at a major international bank, trained as a tax lawyer, and I build and research AI systems across retrieval for law and tax, privacy, computer vision, knowledge graphs, model fine-tuning, quantitative finance, drug-safety data and interpretability, and on nature: species identity in images, land cover from aerial imagery and hiking routes. I have written Python for ten years, including NLP, machine learning and trading strategies.

LinkedIn · Google Scholar · Semantic Scholar · GitHub · ORCID · Contact

Currently working on

Law and tax retrieval: typed negative evidence for jurisdiction and version mismatches (paper), and the engine behind it: hybrid, version- and jurisdiction-aware search over millions of legal documents, citation graphs and a nightly-refreshed corpus
Confidential-document LLM use: in-browser pseudonymisation, my own multilingual NER in ONNX, homomorphic-encryption checks, a blind server model that never sees letters, and how to evaluate what leaves the device
Swiss aerial land cover and cattle-terrace (terracette) detection for erosion: terrain and LiDAR features, per-class calibration, schema-validated output that carries its evidence
Evidence-linked knowledge graphs and automated LLM content pipelines
Animals and plants in image generation: getting the species right, for about 232,000 European taxa, with diffusion fine-tuning (LoRA, DreamBooth, textual inversion) and a BioCLIP and vision-language check
Fine-tuning of language models (LoRA on 0.6 to 9 billion parameters), encoders (multilingual NER, machine-written-text detection) and evaluation of local LLMs
A self-supervised financial model on trade and quote data, against HAR-RV baselines
Drug-safety signal detection on twelve quarters of FDA adverse-event reports: disproportionality statistics, label versioning and a date-aware signal-to-label check
Interpretability: logit lens, ablations and coefficient swaps on an open language model

Publications

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