Research Scientist

Traverse; San Francisco (CA); On-site; USD 140,000 - 230,000; Full time

Job summary

Traverse is a research data lab building reinforcement learning environments for frontier AI labs. As a Research Scientist, you will design and build RL environments that teach models to do work that has historically required years of human expertise.

You’ll work at the boundary of research and domain knowledge, turning that into training signals that actually work. We value rigorous reasoning about hard problems and fast learning.

Qualifications

• MS, PhD, or equivalent depth of experience in a rigorous field.
• Strong implementation skills and ability to run experiments quickly.
• Genuine curiosity about domains outside your own and how expertise works in them.
• You go deep on problems and don't stop at the obvious answer.

Responsibilities

• Research and develop novel approaches to reward modeling, environment design, and evaluation for non-deterministic domains.
• Collaborate with domain experts to understand mastery and translate it into training signals.
• Build and run experiments to validate environment improvements in model capabilities.
• Contribute to Traverse's research output and help establish methodology across new verticals.
• Work directly with partner labs to integrate environments into post-training pipelines.

Skills

Research

Experimentation

Problem solving

Deep reasoning

Education

MS/PhD or equivalent depth in a rigorous field

Tools

Python

Job description

Traverse is a research data lab building reinforcement learning environments for frontier AI labs. We focus on the non-deterministic, taste-dependent work that makes up most of the economy and that nobody else has figured out how to train models on. We work directly with the labs building the most capable models on earth as a thought partner. Backed by Y Combinator.

About the Role

As a Research Scientist, you will design and build RL environments that teach models to do work that has historically required years of human expertise. You'll work at the boundary of research and domain knowledge, figuring out how to formalize what good performance looks like in messy, real-world domains and turning that into training signal that actually works.We're looking for people who think carefully about what makes a good environment, not just what makes a functional one. No prior ML or AI experience is required - we care about the ability to reason rigorously about hard problems and learn fast.

In this role, you will

• Research and develop novel approaches to reward modeling, environment design, and evaluation for non-deterministic domains
• Collaborate with domain experts to understand what mastery looks like in a given field and translate that into training signal
• Build and run experiments to validate that environments actually improve model capabilities
• Contribute to Traverse's research output and help establish our methodology across new verticals
• Work directly with partner labs to integrate environments into their post-training pipelines

Your background looks something like this

• MS, PhD, or equivalent depth of experience in any rigorous field
• Strong implementation skills and ability to run experiments quickly
• Genuine curiosity about domains outside your own and how expertise works in them
• You go deep on problems and don't stop at the obvious answer

Bonus

• Experience with reinforcement learning, RLHF, reward modeling, or LLM evaluation