About

Hey there! I am a Researcher at JetBrains, working on post-training code models and agents with supervised fine-tuning (SFT), distillation, and reinforcement learning (RL), alongside mid-training on agent trajectories.

I work across the research and implementation: framing hypotheses, designing experiments, building trajectory and distillation datasets, and running model training. I currently lead a proof of concept on training a code-understanding subagent with RL.

My research background is in long-horizon coding agents. In my first-author paper, The Complexity Trap (NeurIPS DL4C 2025), we showed that simple observation masking matches LLM summarization on solve rate. Compared with the baseline agent, observation masking reduced cost on SWE-bench Verified by over 60% while maintaining task performance. I am particularly interested in taking this problem into the model: training agents to manage their own context and tool use.

Before research, I worked as a full-stack software engineer and data analyst. That experience still shapes my work: I tend to own the infrastructure as well as the experiments, and draw research questions from practical developer workflows. I completed my master’s degree in computer science with distinction at the Technical University of Munich in November 2025.

If my research interests or projects resonate with you, feel free to reach out or explore my projects and writing below.

• Explore my training work and research on the projects page.
• Check out my invited talks.
• Read my notes and write-ups on the blog.