CTO at Icaro Lab, working across AI safety, model evaluation, and research engineering.
I build infrastructure for studying how AI models and agents behave in practice: their capabilities, failure modes, interactions, and effects on work and organizations.
- MASE: infrastructure for controlled multi-agent experiments.
- Adversarial Humanities Benchmark: evaluating refusal robustness under adversarial reformulations.
- Ora et Labora: repo-first workflows for coding and research agents.
- Ars Operandi: operational adapters and safety boundaries for agentic workflows.



