“One must still have chaos in oneself to be able to give birth to a dancing star.” — Friedrich Nietzsche
MSc Computer Science, ETH Zürich · MATS Scholar
I'm a second-year Master's student in Computer Science at ETH Zürich. I carried out my Master's thesis at MIT and I'm currently a MATS scholar in the LawZero stream, the AI safety organization founded by Yoshua Bengio. My research sits at the intersection of behaviour, interpretability and alignment: I study how human cognition, behaviour and psychology shape modern AI models and agents, and how these agents reason and act within the complex frameworks that are second nature to humans yet remain largely unexplored for AI. I aim to develop systems whose internal reasoning we can interpret and whose behaviour stays reliably aligned with deeply human social processes, while ensuring an efficient and transparent use of resources.
I received my Bachelor's degree in Mathematical and Computing Sciences for Artificial Intelligence from Bocconi University.
I'm always happy to chat about research and open to collaborations, so feel free to reach out!
Other interests: sports, music, art, philosophy
A committed adversarial minority can drive a population of LLM agents into a collective decision none of its members would make alone; we show this is predictable in advance, only partly defendable, and reversible once the adversaries are removed.
Probing whether large language models can identify which intermediate steps causally drive their final answer in chain-of-thought reasoning.
Even in a highly constrained setting where a neutral student model generates unbiased completions, a biased judge can transmit behavioural traits to it through binary preference labels alone.
A nudge seen by a single subagent on a real web page can cascade into the final output of a hierarchical computer-use system, showing the system is only as robust as the subagents that read the web, despite the stronger model that controls them.
Completed as my Master's thesis, supervised by Manuel Cherep, Nikhil Singh, Prof. Pattie Maes, Frédéric Berdoz and Prof. Roger Wattenhofer.
Open-set neural implicit encoding for semantic SLAM, building dense scene representations that jointly capture geometry and open-vocabulary semantics.
Study of the dynamics of random recurrent neural networks, focusing on numerical analysis of dynamical properties such as chaos, stability, and convergence. Developed biologically motivated extensions to explore the intersection of AI and neuroscience for more sustainable and efficient architectures.
Designed and trained neural networks to solve visual jigsaw puzzles by reordering shuffled image patches. Compared CNN-based spatial prediction with pre-trained AlexNet and Vision Transformer encoders combined with Sinkhorn-based permutation optimization.
Fine-tuning transformer-based models for single-cell RNA sequencing data to predict cellular exposure to endocrine-disrupting chemicals and hormonal pathways, linked to neurodevelopmental disorders.
Analysis of the Allen Brain Observatory to model neural dynamics in the mouse visual cortex using correlation analysis, maximum entropy modeling, and classical machine learning for firing rate prediction.
An applied machine learning and deep learning study on gene expression data from breast cancer cells to predict hypoxic and normoxic states, aiming to identify mechanisms linked to treatment resistance.
Analysis of large-scale datasets containing millions of entries to uncover structural patterns and leader–follower dynamics within the Venture Capital ecosystem, quantify and interpret complex network relationships.
Developed Python simulations of the 3D Ising model to analyze magnetization–temperature relationships across varying spin configurations.
A statistical study examining how socio-economic factors and the COVID-19 pandemic relate to the prevalence of mental health disorders across countries.