Research Engineer, Real Environments

San Franciscofull timeOn Site

Apply on Employer Site

mid

$160k–$500k

H1B: no

Mercor The Research Engineer, Real Environments at Mercor builds high-fidelity reinforcement learning environments for AI evaluation and training using enterprise data. The role involves shipping models for workflow extraction, classification, and grading, and automating environment production.

Responsibilities

You’ll work with large enterprises to capture their data and transform it into high-fidelity RL environments for capability evaluations and training datasets for frontier labs. We focus on pushing the frontier of world-building, verifier engineering, and more alongside our partners.
Your goal will be to automate the process of building evals for real work in the economy.
Ship models for workflow extraction, classification, and grading.
Engineer autonomous task refinement processes which distill data taste into pipelines.
Deliver data to customers and deploy into real engagements.
Help define the future of agentic transformation for enterprises around the world.
Deeply learn about the intricacies of enterprises through building evaluations for all aspects of work.
Build end-to-end environments for labs & enterprises by platformizing sandbox app clones, load real data into the sandboxes, build prompts from real workflows, and write verifiers leveraging enterprise expertise & golden outputs.
Systematize the production of environments to scale throughput while maintaining high-quality worlds and verifiers.

Qualification

Experience with Temporal

Required

Prior experience shipping environments – you’ve contributed to an OSS framework, built environments at previous companies, or worked on agentic evaluations.
Strong full-stack engineering skills – you’ll be responsible for everything from infrastructure to app code to analytics
Bias to action – this team is focused on shipping evals, not just philosophizing about them.
Curiosity – being biased towards understanding and digging deep into model behavior and actually looking at the data.
Sweat the details that make a simulation indistinguishable from the real thing and have systems-level thinking skills that allow you to scale up quality.

Preferred

Experience with Temporal, Modal, or similar orchestration/compute services
Experience with synthetic data generation for frontier models.Past work auditing and scrutinizing industry-standard evaluations