Staff Research Engineer — Post-Training & AI Agents

NextToken; Mountain View (CA); On-site; USD 140,000 - 210,000; Full time

Job summary

NextToken is shaping specialized models and the systems that deploy them in Mountain View. You’ll develop data, evaluations, and post-training methods that improve models on real product tasks, understanding where agents fail and carrying experiments to useful deployments.

You’ll work with the founders, systems engineers, and customer teams to design experiments, run training recipes, and turn findings into reusable methods and tools that scale across customer products.

Qualifications

• Understanding of modern post-training methods including RLVR and policy optimization approaches such as GRPO and OPSD.
• Familiarity with frameworks such as PyTorch or JAX for model training and researching code.
• Experience investigating model training, evaluations, agent environments, or related ML research.
• Strong experimental judgment with baselines, controlled comparisons, and attention to generalization and rewards.
• Ability to turn open-ended problems into research plans and explain decisions clearly to teammates and customers.

Responsibilities

• Develop post-training methods for language models and agents, using supervised fine-tuning, RL, and distillation.
• Research synthetic data generation and curation to produce useful learning signals.
• Develop agent environments, reward models, and verifiers, and establish meaningful benchmarks for improvements.
• Work with systems engineers to improve experiment efficiency and address rollout cost and training stability.
• Translate research findings into reusable tools and applications for customer use.
• Communicate findings via technical writing, publications, and open-source contributions.

Skills

RLVR understanding

PyTorch/JAX familiarity

ML research experience

Experiment design

Communication of results

Tools

PyTorch

JAX

Job description

NextToken is shaping specialized models and the systems that deploy them in Mountain View. You’ll develop data, evaluations, and post-training methods that improve models on real product tasks, understanding where agents fail and carrying experiments to useful deployments.

You’ll work with the founders, systems engineers, and customer teams to design experiments, run training recipes, and turn findings into reusable methods and tools that scale across customer products.