Maria Rafaela Gkeka is joining NOFire AI as a Member of Technical Staff on the Runtime & Controls team. She is a systems researcher, still finishing a PhD at the Computer Systems Lab at the University of Thessaly, and she joins from Nubificus, the systems company founded by NOFire AI co-founder and Chief Scientist Anastasios Nanos.
Her work here starts from an uncomfortable premise: the policy that governs an agent should not be a document somebody wrote by hand and hoped stayed true. It should be derived from what the agent actually did. That is what she is building, and it carries a trade-off she is unusually willing to state out loud.
Background
Maria Rafaela has been at the University of Thessaly since 2013. A diploma in Electrical and Computer Engineering, then a Master's in Science and Technology of Electrical and Computer Engineering, and since 2018 a PhD at the Computer Systems Lab, which she is still finishing. Eight years into research, the thing that keeps her there is the seam between software and hardware. Compilers especially.
She joined Nubificus as a Systems Researcher in December 2024 and stayed until this May. That makes her the sixth person now at NOFire AI to have come out of that company, after Charalampos Mainas, Kostis Papazafeiropoulos, Panagiotis Mavrikos, Anastasia Mallikopoulou and Maria Gkoutha.
Her own description of what she enjoys is getting more out of a system than it looks like it has to give. Profiling an agent's behaviour precisely enough to derive a policy from it is that same instinct pointed at an unfamiliar target: the scarce resource is not memory or cycles, but certainty about what a program meant to do.
What Maria Rafaela will own
On the Runtime & Controls team, Maria Rafaela works on the Profiler: the component that observes what an agent actually did and turns that recorded behaviour into the policy that constrains it, instead of relying on a rulebook maintained by hand. The discipline in it is the refusal to over-claim. Only what the system can genuinely vouch for gets allowed through.
Underneath sits a real tension, and she is direct about it. Watching an agent from outside its sandbox produces a verdict nobody can forge, because the observation does not depend on the agent's cooperation. That same distance costs you insight into intent: from outside you see what happened, not what was meant. How much of that understanding can be won back, and which parts honestly cannot be, is the question her work turns on.
More about Maria Rafaela
How did you get into systems research?
I studied Electrical and Computer Engineering, and what really pulled me in was the space between software and hardware, compilers especially. I liked that way of thinking. I also just enjoy solving problems, especially the kind where you're trying to get more out of a system than it seems to have to give.
Why NOFire AI?
AI agents are only as safe as the boundary around them. That's the kind of problem I enjoy working on, concrete, and hard to fake. NOFire AI treats it as something you build and observe, not just a policy you write and hope holds. It felt like the right place to keep doing that kind of work.
What will you be working on here?
I work on the Profiler: turning what an agent actually did into the policy that governs it, instead of a rulebook someone writes by hand and hopes stays accurate. Concretely, that means looking at recorded behaviour and being honest about how well we actually understood it, then only allowing what we can genuinely vouch for.
The deeper trade-off is this: watching an agent from outside its sandbox gives you a verdict nobody can fake, but it costs you some understanding of what the agent actually meant to do. My job is figuring out how much of that understanding we can win back, and being upfront about the parts we can't.
Tools you rely on?
Linux and C at heart, though I went through a Windows phase along the way. Whatever else changed, the terminal stayed constant.
Outside of work?
Time outdoors and the quiet that comes with it, and books that nudge my thinking sideways or teach me something new. I also volunteer regularly.
Maria Rafaela is based in Greece, and joins the team whose job is to make an agent's limits provable rather than declared.
