Qualifications
Key Responsibilities:
Design, implement, and maintain SFT and RL post-training pipelines for multi-step coding agents.
Train and adapt large language models (LLMs) for agent workflows, focusing on planning, tool utilization, and multi-step interactions within JetBrains IDEs.
Develop and construct evaluation and simulation environments for coding agents to operate, be assessed, and compared on realistic developer tasks.
Create evaluation frameworks and metrics to analyze agent behavior, scrutinize logs, and iteratively refine training, data, and reward design.
Interpret training and evaluation outcomes to suggest and implement enhancements to model architectures, training strategies, and datasets.
Engage with large-scale infrastructure, including distributed training on GPU clusters and extensive MapReduce-style data processing to support pre-training and fine-tuning datasets.
Collaborate closely with research, product, and infrastructure teams to translate high-level product visions into tangible models, experiments, and features.
Qualifications:
Demonstrated experience in training LLMs (pre-training, fine-tuning, or post-training) in either a research or production environment.
Proficiency with modern deep learning frameworks, such as PyTorch, and specialized LLM training stacks (e.g., Megatron, NeMo, or similar).
Strong foundational knowledge of LLM training principles, including tokenization, data pipelines, batching, mixed precision, distributed training, and troubleshooting unstable runs.
Capability to lead projects from inception to completion, navigating from high-level challenges to design, experimentation, implementation, and iterative improvement.
A product-centric mindset; an understanding of how agents are utilized by developers and the ability to translate product needs and potential challenges into modeling and evaluation tasks.
At least 3 years of relevant experience in the field.
About the job
At JetBrains, we are passionate about crafting exceptional developer tools. Since our inception in 2000, we have dedicated ourselves to creating the most effective and innovative solutions for developers around the globe. With the rise of AI-assisted technology, we are at the forefront of integrating these advancements into our integrated development environments (IDEs).
Join our Agentic Models team as a Research Engineer, where you will play a pivotal role in developing multi-step coding agents capable of comprehending extensive codebases, strategizing modifications, utilizing tools, and engaging interactively with users. You will be responsible for designing models, training loops, and evaluation pipelines that drive these intelligent coding agents.
Your work will involve merging supervised fine-tuning (SFT) techniques with reinforcement learning (RL) methodologies, coupled with product-oriented evaluations, leveraging our distributed GPU and MapReduce clusters to deploy models across JetBrains products.
About JetBrains s.r.o.
JetBrains is a global leader in the development of intelligent software solutions, committed to advancing the productivity of developers worldwide through innovative tools. Our focus on quality and user experience has made us a trusted partner for software development teams across various industries.
