Job Description

Want to build the simulated worlds that test what frontier models are really capable of?

This is a chance to join a team advancing the science of post-training and scalable evaluation — building reinforcement learning environments that push reasoning, planning, and long-horizon behaviour to their limits.

Instead of static benchmarks, you’ll create dynamic simulations that measure real intelligence — not just accuracy. You’ll design new post-training algorithms (RLHF, DPO, GRPO and beyond), develop richer reward models that move past exact-match scoring, and build evaluation frameworks that define how next-generation AI is trained, aligned, and understood.

The work combines deep research with hands-on implementation — from writing papers to seeing your methods deployed in live systems. It’s ideal for researchers who care about bridging academic insight and practical impact, helping AI progress beyond metrics that no longer tell the whole story.

You’ll bring:

• Research experience in post-training, reinforcement learning, or evaluation for LLMs.

• Strong understanding of transformer models and experimental design.

• Publication record at leading venues (NeurIPS, ICLR, ICML, ACL, EMNLP).

• PhD or equivalent research experience in CS, ML, NLP, or RL.

Package: Up to $300K base (DOE) + meaningful equity + comprehensive benefits (401k, unlimited PTO, relocation and sponsorship available).
Location: On-site in New York (preferred).

If you want to shape how AI is trained, tested, and trusted — this is the place to do it.
All applicants will receive a response.