jacopo.png

  Trieste, Italy

Jacopo Zacchigna

Master's student in Data Science & AI · SPAR Research Fellow (Spring 2026)

I’m a final-year Master’s student in Data Science and Artificial Intelligence at the University of Trieste, focused on deep learning and mechanistic interpretability.

I want to understand how LLMs work internally and how we can use that knowledge to build better and safer systems for the future. I’m open to collaborating on research in these areas. Feel free to reach out.


Recent work · SPAR Research Fellow (Feb–May 2026) on Implicit Personalization. I worked on probing internal user representations in LLMs (persona vectors). See the project page for details.

In parallel, I have been working on mechanistic interpretability of Mixture-of-Experts models (OLMoE, GPT-OSS), focusing on adapting HeadPursuit and SOMP to study expert specialization.


Background · B.S. and M.S. at the University of Trieste (2021–2026) · AI Consultant at Obloo (2024–2025) · Published in Plants

Tooling · PyTorch · NNsight / nnterp · Hugging Face · CUDA · OpenMP · Slurm · Python · C/C++

🇮🇹 Italian (native) · 🇬🇧 English (C1)

selected publications

  1. Toward automated extraction of metadata from herbarium specimen labels
    Jacopo Zacchigna and others
    Plants, 2026