Have you built the environments and benchmarks that AI models learn from? Then this is your seat. A seed-stage AI company in San Francisco partners closely with the leading AI labs, creating the reinforcement learning tasks, environments and benchmarks those labs use to train and test their models. You'd be their first engineer in San Francisco, and from about month six you'd lead your own team. What you'll own Designing and building RL tasks and environments that frontier models train on Eval harnesses and benchmarks that show whether a model actually got better Post-training experiments, and turning what you learn into better environments Hiring and leading a team in San Francisco as the company grows Must-haves You've built RL environments, tasks, benchmarks or evaluation harnesses with your own hands. Only operating an existing eval pipeline won't be enough here. Hands-on post-training experience (SFT, DPO, GRPO, RLHF or similar), ideally with a result you can point to, such as a benchmark score that moved Able to work full time from the San Francisco office now, on a 9-9-6 schedule Roughly 1–5 years of experience Bonus points A benchmark or post-training project of your own, public on GitHub Experience at a data or RL-environment company, or on a post-training team at an AI lab or strong university lab You're outside San Francisco but genuinely AI-obsessed and ready to move What's in it for you $150k–$200k base, plus equity (0.25–0.5%) On top of that, 4% of project profits, which can realistically bring total pay to around $500k a year A founding role working side by side with frontier labs Good to know Full-time, on-site in San Francisco, 9-9-6 Sponsorship is limited: please tell us your exact work authorization up front Process: founder call - culture conversation with the co-founder - take-home build (RL and evals) - one day on site - offer
