You build data pipelines yourself and you love digging into individual examples. Turn real expert work into training environments for frontier AI.
An early-stage San Francisco startup converts real professional work, such as an engineering drawing and its 3D model, into reinforcement-learning tasks and evaluations for frontier AI labs and enterprises. It starts with engineering and life sciences, already runs at an eight-figure revenue rate, and works with several of the leading labs. The team is six people today and heading to about fifteen by year end.
They're hiring 2 engineers for this team.
What you'll own
- Pipelines that turn raw professional work into RL tasks and evals
- Dataset definitions, worked out together with lab researchers
- Quality checks and graders, designed with domain experts
- Spotting ambiguous tasks and tasks a model could game, before they ship
Must-haves
- Hands-on work in data delivery, RL environments or evals at a data company, or a background in quant trading (internships included). Working at a data company in an unrelated function, like payments, doesn't qualify.
- A strong programmer who built pipelines, automation or AI applications personally, not only specified them
- At least 1 year of experience
- Able to start soon
- Working from the San Francisco office, 9-9-6, and happy in a startup where things shift quickly
Bonus points
- Experience with synthetic data, agents or evaluating models
- A data-quality problem you tracked down by inspecting individual examples
- Side projects you shipped on your own time
What's in it for you
- $200k–$400k base + 0.1–0.5% equity
- Working day to day with AI researchers, in a small and very strong team
- Health, dental and vision cover, gym access, and $1,000 a month for meals
Good to know
- Full-time, on-site in San Francisco, 9-9-6
- Visa options are limited: US citizens and green card holders, refugees or asylees, and H-1B, O-1 or TN holders can be considered, so please share your exact status
- Process: profile review - system-design interview (1 hour) - open-ended coding interview (1 hour, AI tools allowed) - one day on-site with a project and a short presentation - offer
#J-18808-Ljbffr