machine learning · production workflows
ABOUT THE ROLE
Join an early-stage AI company building post-training methods and infrastructure for production AI agents. In this research engineering role, you will connect machine learning research with practical systems, helping improve model quality, cost, and latency in real-world workflows.
WHAT YOU'LL DO
- Build evaluation suites, graders, and data pipelines using production workflows and traces.
- Lead post-training work, including data preparation, evaluations, reinforcement learning environments, and training runs.
- Implement agent environments and harnesses for reproducible experiments and training data generation.
- Run and analyze post-training experiments, reproduce promising research, and debug related systems.
- Turn successful experiments into reliable components for post-training and continual learning systems.
WHAT WE'RE LOOKING FOR
- One to five years of engineering experience at an AI startup or in an ML research group, working hands-on with large language models in post-training, evaluation, or inference.
- Strong programming skills, especially in Python, and solid machine learning and computer science fundamentals.
- Hands-on experience training, evaluating, and serving LLMs, beyond using model APIs or prompt and agent wrappers.
- Experience with tools and systems such as PyTorch, JAX, LLM training, evaluation harnesses, model serving, vLLM, SGLang, Megatron, or reinforcement learning environments.
- A shipped system, engineering work, open-source contribution, or substantive research project that demonstrates technical depth