SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Baseten powers mission-critical inference for leading AI companies including Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma, and Writer. The company recently raised $1.5B in Series F funding and provides applied AI research, flexible infrastructure, and developer tooling to help frontier AI companies bring cutting-edge models into production.
In this role, you will work at the applied end of post-training research, collaborating directly with stakeholders from the world's fastest-growing AI companies. Your primary focus is post-training open-source models to outperform frontier closed models on specialized tasks. Day-to-day responsibilities include extracting signal from complex, domain-specific datasets and building reward functions, environments, evaluation harnesses, and training pipelines that translate that signal into better models. The models you train ship to production and reach millions of users.
Key responsibilities include:
- Designing and running post-training pipelines: SFT, GRPO, DPO, RLVR, reward function engineering, and synthetic data generation
- Building task-specific training environments and evaluations tailored to customer domains such as healthcare, code generation, and legal, spanning multi-turn tool use, sandboxed execution, and agentic workflows
- Working directly with customers to translate production data into training signal, designing reward loops from real usage patterns and handling distribution shift
- Running and analyzing training experiments end-to-end: diagnosing reward hacking, importance sampling drift, and advantage estimation instabilities
- Publishing findings at top venues and contributing to Baseten's open-source training libraries
The ideal candidate is excited by the prospect of shipping models into production, can translate domain-specific requirements into effective training curricula, and knows when to be rigorous and when to iterate fast.
REQUIREMENTS:
- Hands-on experience training LLMs with reinforcement learning, with demonstrated understanding of GRPO or PPO beyond recipe-level reproduction, including group advantage computation, clipped objectives, and KL penalty design
- Strong intuition for reward engineering: ability to distinguish between a reward that trains effectively and one that will exploit at scale
- Experience building multi-turn agent environments with tool use, not limited to single-turn question-answering setups
- Comfort working across the full pipeline from dataset construction through training, evaluation, and deployment
- Experience with production ML systems, with preference for candidates who have closed a training–inference loop where production data feeds back into model improvement
PREFERRED QUALIFICATIONS:
- Experience with RL training frameworks
- Publications at NeurIPS, ICML, ICLR focused on RL for LLMs, reward modeling, or alignment
