Hi, I’m Özgür.

I Evaluate How Frontier AI Behaves in the Real World

I have roots on product strategy, UX and devleopment and I am AI behaviour researcher working at the intersection of model evaluation, system design, and sociotechnical risk. My work focuses on how AI systems actually behave once they interact with users, incentives, interfaces, and organisational workflows. I turn complex behavioural patterns into clear metrics that product teams and researchers can act on.

Research Interest

• detecting emergent situational awareness’.
• identifying deceptive or strategically selective model behaviours’.
• stress-testing model risk attitudes under uncertainty’.
• evaluating content provenance and model handling of user-owned inputs’.

I use DeepEval, OpenAI Evals, and custom behavioural experiments—some adapted from recent cognitive neuroscience decision-making studies—to probe model behaviour beyond surface-level benchmarks.

My Mission

To build the next generation of AI evaluation practices’.
Ones that capture real-world behaviour, expose hidden risks, and support responsible innovation as frontier models grow more capable and more embedded in society.

Model evaluation for Single Label Classification Tasks with Transformers Model

The overall goal of the intent classification task is to develop a machine learning model that accurately identifies the intent behind natural language statements within the enteral tube feeding domain. The objective is to improve the efficiency and effectiveness of customer service interactions by automatically routing input sentences to corresponding flow’s dialogue initiator.

Mapping the Hidden Networks of AI Harm

Project investigates how opaque AI systems generate real-world harms that individuals cannot contest, and connects these failures to gaps in the enforcement of the EU AI Act. By analyzing a curated database of 1,000+ AI incidents and 3,200+ entities, I have traced patterns of opacity in hiring, healthcare, and content moderation, revealing how corporate practices collide with regulatory obligations for transparency, risk management, and bias prevention.

Foresight Explorer Tool

multidimensional framework for curating, classifying, and mapping weak signals from emerging technologies and sociopolitical shifts. It helps trace how technical, cultural, and systemic changes interact

Github Profile

I am also an expert evaluator and mentor for the EU-funded programs. Experienced in evaluating 20+ proposals for EU-funded programs, including Horizon Europe (HE), European Research Council (ERC), EIT, EIT Culture & Creativity. Expertise in product design, audiovisual media, cultural heritage, and gaming with a focus on innovation, education, and business creation. I have offered a qualitative and quantitative analysis along with sufficient rationale to support the scores given to each criterion.

If you’re building something ambitious and seek critical insight with warm collaboration, let’s connect.

Education:

Product Design, MDes
2006, Izmir Institute of Technology, İzmir

Computer Science Engineering, BSc
2001, Dokuz Eylül University, İzmir

University City Science Center, USA
Student Exchange Trainee IAESTE, Philadelphia
2000, In DRaW Computing Associates, Inc. I have developed 3D computer graphics software to be used in NASA’s VR trainings