Tech Stack

Description

You will design and implement training pipelines for multi-step coding agents, train LLMs for agent workflows, and build evaluation frameworks to improve model performance. Your work will directly impact developer tools used by millions.

Requirements

• Extensive hands-on experience training LLMs in research or production
• Deep expertise in PyTorch and LLM training stacks (e.g., Megatron, NeMo, verl)
• Strong understanding of LLM architectures, tokenization, data pipelines, distributed training
• Ability to own projects end to end from problem to implementation
• At least 3 years of Python experience in modern ML codebases

Responsibilities

• Design and maintain SFT and RL post-training pipelines for multi-step coding agents
• Train and adapt LLMs for agent workflows, including planning and tool use
• Build evaluation and simulation environments for coding agents
• Develop evaluation frameworks and metrics for agent behavior
• Analyze training results and improve models, architectures, and datasets