Researcher at Laboratory of Data Engineering

I’m interested in mechanistic interpretability, specifically controlled steering of large language models, used to reliably enforce safety, truthfulness, and other desired behavioral properties. Previously, I worked on music generation and source separation with diffusion models and transformers, as well as vector quantile regression through neural optimal transport. I’m trained as a mathematician and computer scientist, with a BSc in Mathematics, an MSc in Computer Science, and a PhD in Computer Science from Sapienza University of Rome.

Research Interests

• Mechanistic Interpretabillity
• Model Steering

Experience & Education

• PostDoc in Mechanistic Interpretability, LADE Lab, Area Science Park, Trieste (May 2025 – ongoing): steering of LLMs
• Researcher in Audio LLMs, GLADIA Lab, Sapienza University of Rome (Mar. 2024 – Jun. 2024): steering of music LLMs and music transcription with large language models
• PhD in Computer Science, Sapienza University of Rome (May 2024)
• M.S. in Computer Science, Sapienza University of Rome (Jan. 2020)
• B.S. in Mathematics, Sapienza University of Rome (Jul. 2017)