About

I am a Ph.D. researcher at the Digital Aviation Research and Technology Centre (DARTeC) at Cranfield University, affiliated with the School of Aerospace, Transport, and Manufacturing (SATM), under the supervision of Prof. Weisi Guo. My work is supported by an EPSRC iCASE award with Thales UK (EP/X52475X/1).

Prior to Cranfield, I received my Diploma from the University of West Attica and worked as an undergraduate intern and later as a Research Associate in the Extreme Robotics Laboratory (ERL) at the University of Birmingham under Dr. Manolis Chiou, where I developed probabilistic methods for variable-autonomy robotic systems with a human in the loop, particularly for operation in hazardous environments.

My research is centred on autonomous and intelligent systems, with a particular interest in how agents acquire, represent, and monitor information while they act. I study settings in which the information required for a task is incomplete, uncertain, or distributed across people, sensors, learned models, and other computational systems, and ask how this information can be exposed to decision-making through interfaces that are observable, modular, and amenable to inspection and intervention.

A recurring theme in my work is that behavioural competence alone tells us surprisingly little about the internal condition of an autonomous system. An agent can remain active without becoming better informed, a learned model can produce plausible behaviour without making the representations or processes supporting that behaviour readily observable. I’m interested in methods that make these hidden or implicit states more explicit — from the runtime condition of information acquisition to the internal representations and signals that shape the behaviour of learned agents.

This places my work at the intersection of agent observability, information acquisition, autonomous decision-making, and AI safety. I’m particularly interested in the boundary between measurement and intervention asking what aspects of an intelligent system can be observed reliably, what those measurements actually tell us, and how they can support monitoring, diagnosis, oversight, or downstream control without unnecessarily entangling the observer with the system being observed.

My research draws on reinforcement learning, large language models, planning, robotics, and ideas from control, state estimation, and representation analysis. Although I don’t treat any one of these as the end goal, I’m interested in the architectures and abstractions that make complex intelligent systems more inspectable, such as interfaces through which information, uncertainty, internal state, and acquisition dynamics can become explicit objects that other components — or human operators — can reason about.

Safety-critical robotics and search-and-rescue provide important motivating settings for this work because failures of information acquisition or interpretation can have immediate consequences. The underlying questions, however, extend to autonomous and agentic AI more generally, including systems that reason over long horizons, interact with external information sources, and increasingly rely on learned internal representations that are difficult to inspect directly.

More broadly, I ask a simple question: how do we know what an intelligent system is doing internally, whether its reasoning or information-gathering process is progressing as intended, and when intervention is warranted?

My aim is to develop methods that make autonomous and AI systems more observable, diagnosable, and trustworthy while they operate.