Konstantinos Pitas

I led the pre-training data collection for the upcoming Apertus models, and currently co-lead the SWE/Terminal/Tool Use expert. I’m passionate about agentic AI and committed to building the world’s best fully open agentic LLM. I’m interested in topics such as scalable RL Gym synthesis under resource constraints, and RL training on heterogenious/long-horizon tasks.

Previously, I worked as a Machine Learning Research Scientist at Cognex Corporation (🇨🇭,🇺🇸), where I delivered industry-leading transformer-based object detection models for inspecting industrial production lines, under embedded system constraints. Before that I was a postdoctoral researcher at Inria Grenoble Rhône-Alpes 🇫🇷, where I collaborated with Julyan Arbel and Emtiyaz Khan (🇯🇵). In 2020, I completed my Ph.D. at EPFL🇨🇭, which I obtained under the supervision of professor Pierre Vandergheynst. I completed my MSc. and BSc. in Electrical and Computer Engineering at the Aristotle University of Thessaloniki 🇬🇷.

I’m very passionate about transparent governance and enjoy analysing various topics on political polling and psephology on kassiope.org (in Greek).