cv

Co-Principal Investigator

Rome AI Safety (RAIS)

Proposer and co-PI, together with Donato Crisostomi, of a 7-person research team funded by Coefficient Giving with $80k.

• Developing disentangled, composable steering vectors for fine-grained behavioral control via white-box interventions in activation space.

Research Intern

Institute of Science and Technology Austria (ISTA)

Research on causality and multimodal foundation models within the Locatello Group (PI: Francesco Locatello).

• Co-first author of an ICLR 2026 Oral (top 1.2%): a framework (theory + library) for unsupervised causal effect discovery in RCTs.

Research Student

GLADIA · Sapienza University of Rome

Research on scalable model merging, interpretability, and ability estimation for large language models (PI: Emanuele Rodolà).

• Scaled evolutionary model merging of LLMs to consumer GPUs and developed ability-estimation theory via Item Response Theory: 50× faster, 2% of the data, 85% performance retained. Two first-author papers at ICML 2025 and ACL 2025.
• Co-first author of an ICLR 2026 paper proving injectivity of LLM hidden representations and presenting the first exact inversion algorithm (5M views on the tweeprint).

Research Student

ETH Zürich (LRE Lab)

Applied causal inference to explain and assess LLM performance on math word problems (PI: Mrinmaya Sachan).

• Worked on applying causal inference techniques to large language models to detect spurious heuristics in reasoning tasks.
• Ran large-scale experiments on high-performance computing clusters.

Research Assistant

Vrije Universiteit Amsterdam · KAI Lab

Mitigated hallucinations in LLMs using knowledge graphs for content planning (PI: Ilaria Tiddi).

• Designed Text-to-Graph and Graph-to-Text methods for content planning.
• Trained and evaluated multiple knowledge-graph-integrated LLMs.

Research Student

Sapienza University of Rome

NLP research in e-justice applications using specialized large language models.

• Developed GPT-2-based writing assistant for legal documents.
• Publication at ECAI 2023 (CICERO project).
• Conducted evaluation with legal professionals.

Technology Assistant

Network Institute VU Amsterdam

Technical support for interdisciplinary projects in XR, VR, and motion capture.

• Worked with Unity, C#, iClone, Motive, and Optitrack.
• Organized and presented at Surf XR-On Tour and Reshaping Work 2022 Conference.

Master of Science

École Polytechnique Fédérale de Lausanne (EPFL)

Data Science

• Coursework focused on the Mathematics of data science

Summer School (3 months)

BlueDot Impact

AI Safety

• Studied sparse autoencoder diffing and transfer; work accepted at the UniReps Workshop @ NeurIPS 2024 (blog post track).

Bachelor of Science

Vrije Universiteit Amsterdam

Artificial Intelligence (Major) & Mathematics (Minor)

• Honours Programme (30 credits of advanced coursework, e.g. PDEs)

Bachelor’s Degree

Sapienza University of Rome

Philosophy

• Focused on formal logic and philosophy of language

KHMW Young Talent Incentive Award

Royal Holland Society of Sciences and Humanities

National award presented to the student with the highest GPA in their degree program at a Dutch research university.

Intelligent Systems Competition — 1st Place

Vrije Universiteit Amsterdam

Ranked 1st out of 270 students by developing an intelligent agent for the card game Schnapsen.