Fabian Degen
MSc student in Advanced Computer Science
University of Oxford
I work on LLM agents and causal reasoning. Right now I'm writing my MSc thesis on counterfactual reasoning in LLMs, supervised by Jialin Yu, Philip Torr and Yarin Gal. I co-first-authored Plan2Map, a multimodal benchmark for reconstructing geospatial boundaries from planning documents (under review at EMNLP 2026), and I'm a contributor to TransformerLens, the mechanistic interpretability library. Before Oxford, I studied computer science at the Technical University of Munich with a minor in medicine.
1Publications
Plan2Map is a 208-case benchmark: given only a UK planning document, a system must reconstruct the machine-readable geospatial boundary it describes. I built GeoPlanAgent, the tool-augmented agent system introduced in the paper — it reads the document, locates the site, registers the map against a national basemap, segments the drawn boundary with a fine-tuned SAM 3, and projects it to geographic coordinates, reaching 0.736 mean IoU where direct VLM baselines reach 0.108.
2Open source
Supported by a Manifund grant, I helped build the TransformerLens 3.0 release, which grew the number of supported models from ~200 to ~9,000 and brought model outputs in line with the Hugging Face reference implementations; my main contributions were the backwards-compatibility layer and weight-level Q/K/V splitting. Smaller contributions: InterpretML.
Languages: German, English; some Spanish and French.