AI Research Scientist

Xterraai; San Francisco (CA); On-site; USD 120,000 - 160,000; Full time

Job summary

Xterraai, based in San Francisco, is seeking research scientists to develop innovative AI systems that reason about complex scientific problems. This role combines research and engineering, allowing you to take ownership from ideation to production.

Locally and remotely, you will focus on reinforcement learning, evaluation, and building robust training infrastructure. Ideal candidates possess strong machine learning fundamentals and experience with reinforcement learning.

Qualifications

• Strong fundamentals in machine learning, with experience training large models.
• Demonstrated experience with reinforcement learning or related fields.
• Ability to work across research and engineering.

Responsibilities

• Designing and training systems using reinforcement learning methods.
• Developing fine-grained supervision over intermediate steps.
• Building robust training pipelines and running large-scale experiments.

Skills

Machine learning fundamentals

Reinforcement learning

Distributed training debugging

Reward modeling

AI alignment techniques

Job description

About Xterra

Xterra is a Khosla Ventures-backed company building AI agents that reason about complex scientific problems. We’re not a wrapper around existing models, we’re training our own foundation models on top of large-scale proprietary datasets. This is a rare intersection of frontier AI and real-world scientific impact. Xterra is still in stealth mode.

The Role

We’re looking for research scientists who want to work at the intersection of research and engineering, developing new methods for teaching AI systems to reason. You’ll own problems end-to-end, from idea through experimentation to production.

We’re hiring across multiple levels (early-career through senior/staff) and will calibrate scope and expectations to match your experience.

What You’ll Work On

• Reinforcement learning: Designing and training systems using RLHF, RLAIF, and reward modeling approaches, applied to scientific hypothesis generation and evaluation.
• Process reward models and verifiers: Developing fine-grained supervision over intermediate steps - not just final answers - so the system learns to reason well, not just get lucky.
• Scalable oversight: Contributing to alignment and oversight research - figuring out how to reliably supervise models on complex scientific tasks where ground truth is expensive, delayed, or ambiguous.
• Infrastructure and experimentation: Building robust training pipelines, running large-scale experiments, and iterating quickly across the research-to-production lifecycle.
• Evaluation: Contributing to meaningful benchmarks and evaluation methods for domain-specific reasoning.

What We’re Looking For

• Strong fundamentals in machine learning, with hands-on experience training large models (LLMs preferred but not required).
• Demonstrated experience with reinforcement learning, ideally applied to language models, but strong RL backgrounds from other domains (robotics, game-playing, scientific discovery) are valued.
• Comfort working across the research-engineering spectrum: you can write a paper and you can debug a distributed training job.
• Familiarity with at least some of: reward modeling, RLHF/RLAIF pipelines, search and planning methods, or AI alignment techniques.
• Publication record is a plus but not a strict requirement, we care more about the quality of your thinking and what you’ve built.

At More Senior Levels, We’d Also Expect

• A track record of identifying and driving high-impact research directions independently.
• Experience mentoring other researchers and influencing technical strategy.
• Deep expertise in one or more of the core technical areas listed above.