Selected Projects

Simplifying German-Swiss Administrative Text

An interactive tool that compares how well different LLMs rewrite Swiss administrative text in plain language.

AML Detection with XGBoost & SHAP

A tool that flags suspicious financial transactions that look like money laundering, and explains why it flagged them.

Human Realignment: An Empirical Study of LLMs as Legal Decision-Aids in Moral Dilemmas

With Christoph Engel and Yoan Hermstrüwer

Co-authored research asking whether AI chatbots make the same moral choices humans do in classic ethical dilemmas (like the trolley problem), and whether they can be nudged toward more human-like judgment.

Learning a Decoding Objective from Human-Generated Text

My Master's thesis: teaching a model to write more human-like text by learning what makes a good word choice, instead of hand-tuning the rules.