Type: Independent contractor.

Duration: Approximately 3 months.

Start: As soon as possible.

Rather than primarily building production applications, you’ll work with coding agents across real-world repositories and assess the quality of their work. You’ll review generated code and agent behavior, determine whether solutions are technically correct, identify failure modes, and create the evaluation signals and feedback used to improve model performance.

Think of the coding agent as another engineer whose work you’re reviewing: Can it understand the task? Did it choose the right approach? Is the resulting code correct, robust, and maintainable? Can you explain precisely where it succeeded or failed?

• Design, refine, and iterate on rubrics and evaluation criteria for preference data.
• Review and label data with a high quality bar, catching subtle issues others might overlook.
• Build and maintain pipelines and infrastructure supporting data generation, collection, and evaluation workflows.
• Synthesize findings from data work into clear write-ups, updates, and recommendations for the team.
• Collaborate closely with researchers and engineers to translate qualitative judgment into scalable processes.

• 5+ years of hands-on software engineering experience.
• Strong proficiency in Python, TypeScript/JavaScript, Go, or another major production language.
• Experience working in substantial real-world codebases.
• Strong code-review skills and technical judgment.
• Ability to clearly explain why an implementation is correct, incorrect, or could be improved.
• Strong written communication.
• Experience using modern LLMs or AI coding tools.

Experience with LLM evaluation, coding agents, RLHF, preference data, rubric design, or post-training is a plus, but not required.

• AI interview (~25 minutes).
• Practical code/AI evaluation exercise (~30 minutes).
• Hiring manager interview (~20 minutes).

The practical exercise focuses on your ability to review and evaluate AI-generated code, not competitive programming or algorithm puzzles.

Work arrangement: Fully remote

Eligible regions: North America

Freelance · Remote · North America (only).

Location: North America (Only).

Compensation: $100-$150/hour, please provide a specific hourly rate expectation.