My name is Arthur Bražinskas (pronounced Bra-[zh]-inskas). I'm an AI researcher and founder interested in a simple question with deep implications: what would it take for AI systems to understand a person not as a series of isolated interactions, but as a dynamic, temporally extended individual — someone with evolving goals, expertise, and context?

I'm currently building a company in stealth around this question, drawing on my research in personalization, user modelling, and post-training. More soon.

Previously, I was a Senior Research Scientist at Google DeepMind, where I founded and led a workstream on user signals for Gemini — modelling user behaviour and preferences, building user simulators for model evaluation, and improving alignment and personalization during distillation and post-training. I also contributed to the Gemma family of open-weight models.

My broader interests sit at the intersection of AI, cognitive science, and the study of mind. See my philosophical blog for essays on this topic.

I am currently building a new AI company in stealth. The work explores how AI systems can move beyond isolated interactions and develop a more persistent understanding of individuals over time. It draws on research in language modelling, personalization, user modelling, cognitive science, evaluation, and post-training. More details will be shared in the future.

I received my PhD in Artificial Intelligence from the University of Edinburgh, supervised by Mirella Lapata and Ivan Titov, working on Bayesian machine learning, reinforcement learning, and neural methods for abstractive opinion summarization. This research has since contributed to Amazon Alexa and e-commerce summarization systems. Before that, I earned an MSc in Artificial Intelligence with distinction from the University of Amsterdam and worked on machine learning at Elsevier, Amazon, and Zalando. My work has been published at ACL, EMNLP, NAACL, and related venues.