I enjoy making things. Here are a selection of projects that I have worked on over the years.
A teacher-student framework that distills symbolic STL robustness kernels into a Transformer encoder via geometric kernel alignment.
A Transformer-based decoder that inverts semantic embeddings of Signal Temporal Logic formulae, enabling interpretable requirement mining.
An ongoing research project investigating over-refusal behaviour in LLMs, when models refuse legitimate queries, and how to measure and mitigate it.