Research Engineer, Synthetic Data
Clera · San Francisco, California, United States · $150K–$250K/yr
Technology, Information and Internet · 2-10 employees
About the role
You will build and maintain end-to-end synthetic data pipelines to create structured training tasks for AI agents. Additionally, you will collaborate with subject-matter experts to design generation methods and develop metrics to evaluate task quality and model performance.
What they look for
Requirements
The role requires 2–4 years of engineering experience with proficiency in Python, Docker, and Linux environments. Candidates should have hands-on experience with synthetic data research and the ability to work effectively in an early-stage, unstructured environment.
Benefits
Full description
About the Role
We're an early-stage AI infrastructure company (11–50 people) building the foundational platform for reinforcement learning environments — the tooling that lets AI labs and businesses encode real-world expertise into scalable training and evaluation pipelines. Our engineering team of ~15 includes Olympiad medalists and published researchers, and we're growing it with Research Engineers who want to do hard, impactful work on synthetic data at the frontier of AI alignment.
In this role, you'll own the synthetic data pipeline end-to-end: transforming domain-specific workflows into structured, realistic, and challenging training tasks for AI agents. You'll work directly with subject-matter experts, design generation and validation systems, and develop the metrics that tell us whether synthetic tasks are actually teaching models what we want.
What You'll Do
- Build and maintain the synthetic data pipeline, turning domain-specific workflows into realistic, structured, and challenging training tasks for AI agents.
- Collaborate with subject-matter experts across professional and technical domains to design high-quality synthetic tasks.
- Design synthetic task generation methods that produce diverse, realistic, and learnable data at scale.
- Build tooling to mutate, validate, and iteratively improve synthetic tasks.
- Analyze model and agent performance on synthetic tasks to understand what they teach and where they break down.
- Develop metrics to quantify synthetic task diversity, realism, learnability, and overall quality.
What We're Looking For
Required
- 2–4 years of relevant engineering experience.
- Proficiency in Python, Docker, and Linux environments.
- Hands-on experience with synthetic data research methods.
- Strong intuition for what makes synthetic data "good" — and an honest understanding of its limitations.
- Demonstrated ability to build synthetic data pipelines end-to-end without a fully prescribed roadmap.
- Experience working with environments, evaluations, and benchmarks.
- Detail-oriented mindset for spotting subtle inconsistencies and edge cases in synthetic data.
- Ability to reason from first principles about task design, scoring functions, and failure modes.
- Comfort thriving in unstructured, early-stage environments where you define the path forward.
- Strong written and verbal communication skills for async, cross-timezone collaboration.
Nice to Have
- Background in reinforcement learning or post-training data for large language models.
- Experience building reward signals, graders, or automated QA systems for agent tasks.
- Prior work at an early-stage AI or ML startup.
Compensation & Benefits
- Salary: $150,000 – $250,000 USD annually, depending on experience.
- Visa sponsorship: Available.
- Equity participation in a well-funded, early-stage AI company.
Location
This is an on-site role based in San Francisco, CA. We work together in person — candidates should be prepared to be in the office regularly. Remote arrangements are not available for this position.
