Founded in 2024 as a Sapienza spinout, we now build Physical AI systems for industrial environments, with a research record published at the top tech conferences worldwide. Rooted in European industry, wired into Silicon Valley.
We collaborate with leading universities, research centers, and industry partners to advance artificial intelligence, and translate that knowledge into deployable systems. Our work spans fundamental research, applied AI, and system-level design, with a strong emphasis on correctness, reliability, and scalability.
Our co-founder works from the Bay Area. That is not a satellite office — it is how we stay inside the conversation that sets the direction of this field months before it reaches a paper.
Rome gives us researchers who go deep and stay on a problem. The Bay Area gives us line of sight to what is about to matter. Work moves in both directions: a method read early in California gets stress-tested in Rome, and a result from Rome gets shown to the people best placed to break it.
Where the lab sits. Shared supervision of PhD researchers, and the origin of the spinout.
Working research relationships, not logos — WIZARD and T-Rex are both co-authored collaborations.
A multi-year relationship with a global industrial technology leader. Project detail stays confidential.
Independent of technologies or market cycles.
We would rather hold a small number of problems and go all the way down than spread thin across many. Technical depth and scientific integrity come before expansion.
If a result would not survive peer review, it does not ship. Everything we build is benchmarked, documented, and honest about what it does not yet do.
We design systems for the environment they will actually run in — long-term capability rather than a working demonstration.
Work is organised around milestones with defined outputs, and reported in plain language you can put straight into an internal update.
We start by establishing what a successful outcome actually looks like, and whether the problem is research-shaped at all. Sometimes the honest answer is that it is not.
Every milestone produces something measurable and an explanation of what it means for the capability being built — not a status report.
A method that only works in simulation is not finished. We validate on real hardware, in the conditions the system will meet.
