I’m an MSc student in Computer Science at ETH Zurich, majoring in AI/ML. My work focuses primarily on the safety, alignment, and evaluation of AI agents. I love building systems, measuring how they behave, and improving their capabilities and efficiency.
Research interests
Background
I interned as a SWE on Amazon’s Alexa AI team, where I built a service to improve NER. I earned a BEng in Computer Engineering from Politecnico di Torino and was selected for its Talents Program. I also received the Agon Scholarship for national and international rowing achievements, including an indoor rowing world record.
News
-
Published Agent Memory Is a Surface for Endogenous Authorization Laundering, studying safety failures that emerge within persistent memory.
-
Released Terminal-Bench 4.0, featured in the launch evaluations for GPT-6 Astra, Claude Fable 5.1 and Mythos 5.1, and Gemini 3.8 Flash.
-
Released Terminal-Bench 3.0, featured in the launch evaluations for Claude Opus 5, Gemini 3.7 Flash, Grok 4.6, and GLM-5.3.
-
Published a study of Linear Attention Architectures.
-
CocoaBench was accepted at COLM 2026.
-
Joined EleutherAI’s SOAR program for summer research on agent memory safety.
-
Open-sourced EveryEvalEver, a datastore with a standardized schema for AI evaluation results.
-
Served as a reviewer for three ICML 2026 workshops on AI safety.
-
Published an article on Safety Evals and TTC, later cited in research from the UK AI Security Institute.
-
Released CocoaBench, a benchmark for unified digital agents.
-
Joined Amazon’s Alexa AI team as a software engineering intern, working on named entity recognition.
Outside research
I enjoy rowing, Muay Thai, weight training, hiking, travelling, and trying new foods (though nothing beats carbonara with Coca-Cola).