Research Engineer – Agentic Models

About the role

JetBrains is building multi‑step coding agents that understand large codebases, plan changes, call tools, and iterate with developers inside our IDEs. As a Research Engineer on the Agentic Models team you will own the models, training loops, and evaluation pipelines that power these agents.

Key responsibilities

• Design, implement, and maintain supervised fine‑tuning (SFT) and reinforcement‑learning‑style post‑training pipelines for multi‑step coding agents.
• Train and adapt large language models (LLMs) for agent workflows, including planning, tool use, and multi‑step interactions within JetBrains IDEs.
• Build evaluation and simulation environments where coding agents can be measured on realistic developer tasks, and create metrics and dashboards to close the loop from evaluation back to training.
• Analyze training and evaluation results to propose improvements to model architectures, training recipes, and datasets.
• Collaborate with research, product, and infrastructure teams to turn high‑level product visions into concrete models, experiments, and shipped features.

Required profile

• 3+ years of hands‑on experience training LLMs (pre‑training, fine‑tuning, or post‑training) in research or production settings.
• Deep expertise with PyTorch and LLM training stacks such as Megatron, NeMo, or similar.
• Strong understanding of LLM fundamentals, distributed GPU training, mixed‑precision, and debugging unstable runs.
• Product‑aware mindset with ability to translate developer‑focused product needs into modeling and evaluation work.

Required skills

• Python programming (clean, maintainable code).
• PyTorch, Megatron, NeMo, verl.
• ML orchestration tools: Kubeflow, Dagster, Airflow, ZenML.
• Job schedulers: Kubernetes, SLURM.
• Large‑scale data pipelines (MapReduce‑style clusters, multi‑node GPU training).
• Experiment tracking and observability: Weights & Biases, MLflow, Langfuse.
• Inference optimization and model serving.