Alexander Kovrigin

Hi! I am a Junior Researcher at JetBrains, working on improving LLMs on SWE tasks using Reinforcement Learning. I am pursuing a Master's degree in Data Science under the supervision of Prof. Dr. Dmitry Vetrov at Constructor University in Bremen, Germany.

I earned a silver medal at ICPC Northwestern Europe Regional Contest (NWERC 2023) and received a first prize award at the International Mathematics Competition for University Students (IMC 2023). During my Bachelor's studies, I worked in a research group under the supervision of Andrey Ustyuzhanin.

Email  /  GitHub  /  Google Scholar  /  LinkedIn  /  CV

profile photo

Research
project image

PIPer: On-Device Environment Setup via Online Reinforcement Learning


Alexander Kovrigin, Aleksandra Eliseeva, Konstantin Grotov, Egor Bogomolov, Yaroslav Zharov
NeurIPS 2025 Fourth Workshop on Deep Learning for Code, 2025
arxiv / code / hf /

We develop a specialized LLM for automated environment setup by combining supervised fine-tuning with Reinforcement Learning with Verifiable Rewards. By evaluating on EnvBench-Python, we show that an 8B model reaches GPT-4o– and Qwen3-32B–level performance, demonstrating the effectiveness of SWE tasks training of consumer-scale models.
project image

EnvBench: A benchmark for automated development environment setup


Aleksandra Eliseeva, Alexander Kovrigin, Ilia Kholkin, Egor Bogomolov, Yaroslav Zharov
ICLR 2025 Third Workshop on Deep Learning for Code, 2025
arxiv / code /

We have collected the largest dataset to date for automated environment setup and introduced a robust framework for developing and evaluating LLM-based agents that tackle environment setup challenges. By benchmarking existing LLM agents, we identified a large gap in the quality between expert-produced and LLM-generated scripts.
project image

Long Code Arena: a Set of Benchmarks for Long-Context Code Models


Egor Bogomolov, Aleksandra Eliseeva, Timur Galimzyanov, Evgeniy Glukhov, Anton Shapkin, Maria Tigina, Yaroslav Golubev, Alexander Kovrigin, Arie van Deursen, Maliheh Izadi, Timofey Bryksin
arXiv, 2024
arxiv / code / hf /

Long Code Arena is a suite of benchmarks for code-related tasks with large contexts, up to a whole code repository. It currently spans six different tasks and contains six datasets.
project image

On The Importance of Reasoning for Context Retrieval in Repository-Level Code Editing


Alexander Kovrigin, Aleksandra Eliseeva, Yaroslav Zharov, Timofey Bryksin
arXiv, 2024
arxiv / code /

We experiment with coding agents for repository-level code editing. Our findings show that more complex reasoning improves the ability to locate relevant code.
project image

Beyond dynamics: learning to discover conservation principles


Antonii Belyshev, Alexander Kovrigin, Andrey Ustyuzhanin
Machine Learning: Science and Technology, 2024
paper /

We introduce a method using representation learning and topology to explore conservation law spaces. It is robust to noise and identifies new conservation principles, including in quantum systems.

Design and source code from Jon Barron's website