Research Engineer

About the role

This is a high-ownership role for someone who wants to work at the frontier of LLM evaluation, agentic systems, and startup building. You will help design and execute projects across coding data, computer-use environments, benchmark development, and research infrastructure.

You will work closely with the founders on both technical execution and company-building. That means this role is not limited to pure research or pure engineering; it sits at the intersection of research, product, operations, and GTM.

What you’ll work on

• Build and improve evaluation tasks, datasets, and environments for frontier models
• Lead portions of our computer-use and coding-data projects end to end
• Create tasks for coding, terminal, and computer-use benchmarks
• Help develop new samples and benchmark ideas across new domains
• Contribute to internal and open-source tooling for running agent evaluations
• Help source novel codebases and datasets, including areas like Fortran, COBOL, and embedded C
• Write technical blog posts or research-style writeups about benchmarks, datasets, and methods
• Support hiring, recruiting, and interviewing as the team grows
• Potentially contribute to papers, open-source benchmarks, and research collaborations as time permits

Example projects

• Building realistic computer-use environments with strong verifiers
• Designing coding and agent-eval datasets that surface capability gaps in frontier models
• Creating benchmarks we can publish publicly as part of Refresh’s research presence
• Automating the creation of tasks, environments, and graders for large-scale data generation
• Exploring adjacent tools and platforms that support trajectory collection, eval creation, and failure-mode discovery

What we’re looking for

• Strong engineering ability and high agency
• Interest in frontier LLMs, agent systems, evals, and research infrastructure
• Comfort working across ambiguity and taking ownership of poorly defined problems
• Strong writing and communication skills
• Ability to move between research, engineering, and operational work as needed
• Excitement about working closely with founders in an early-stage startup environment

Nice to have

• Experience with LLM evals, benchmarks, or training data
• Experience building agents, developer tools, or research infrastructure
• Familiarity with modern startup engineering stacks
• Interest in open-source research tooling and benchmark design
• Interest in publishing technical writing or papers

What you’ll learn

• How frontier model evals and data pipelines are built in practice
• Early-stage startup operations, scaling, and execution
• The modern startup stack across infrastructure, product, and deployment
• LLM training fundamentals, research workflows, and data operations
• How to build systems end to end
• How to lead projects and help grow a team