Founding Engineer, RL Environments and Evals (Seed-Stage AI)

A seed-stage AI startup that supplies training environments and benchmarks to frontier model labs is hiring founding engineers to work in person with its two founders. The job is to turn real agent failures into rigorous environments that labs will actually adopt.

What you’ll own

Build synthetic worlds from the ground up, such as an entire simulated organization with generated emails, records and files, where an agent can spend weeks working

Take lab deliverables from scoping through shipping, without needing close supervision

Study production agent traces to find genuine failure patterns, and recreate those as new benchmarks and environments

Write graders that remain trustworthy even if the agent outperforms any person checking its work, and make them resistant to reward hacking

Simulate tools such as email and CRM systems faithfully enough to hold up under an agent’s tool use

Validate that benchmark rankings reflect how agents behave in production

Profile

One to five years in engineering or research, with exceptional new grads from labs or research groups also considered

Prior work building evals, benchmarks, RL environments or training datasets for LLMs or agents

A background at an AI lab, AI data company, agent startup or leading research group

A degree in computer science, math or a related field from a top program

Python skills

Pluses: exposing a flaw in a popular benchmark, catching reward hacking, starting a company, or placing well in math, coding, debate or game competitions

On-site in the San Francisco Bay Area. Visa transfers considered.