Applied Researcher

Confidential; New York (NY); Hybrid; USD 120,000 - 180,000; Full time

Job summary

Confidential is hiring for a hybrid, US-based research role focused on reinforcement learning in collaboration with frontier AI labs. You will be hands-on designing environments, datasets and experiments that advance model capability, with emphasis on RL training data and post-training methods.

This is a research-heavy position joining two applied researchers to meet fast-growing customer demand. Sponsorship is available, and the team values deep RL understanding over pure engineering execution.

Qualifications

• 2+ years of experience as an applied researcher or research scientist.
• Hands-on RL work at a frontier lab or equivalent with demonstrable results.
• Deep understanding of RL and training data beyond tooling.
• Experience with post-training methods such as GRPO, preference optimization and SFT.

Responsibilities

• Collaborate with customer research labs to translate goals into concrete environments, datasets and experiments.
• Design and run post-training loops and identify signals that improve model performance.
• Shape high‑fidelity trajectories and training data with attention to correctness and edge cases.
• Build reward functions, validators and eval harnesses to resist reward hacking and show capability gains.
• Analyze results across distributed GPU clusters to decide next steps.

Skills

Applied researcher

Reinforcement learning

RL training data

Post-training methods

Research focus

Tools

GRPO

SFT

Preference optimization

Job description

Location: Hybrid, New York City or Seattle (3 days/week in-office)

Sponsorship: Available, including new H1B, H1B transfers, OPT and TN.

I’m recruiting on behalf of a confidential client, a well-funded, 15-person AI startup that works directly with frontier AI labs on their hardest post‑training problems. You’ll be hands‑on in reinforcement learning, designing the environments, data and experiments that move model capability.

This is a research role first. The team cares most about how deeply you understand RL and training data, and less about pure engineering execution. You’ll join the two applied researchers already on the team to keep up with fast‑growing customer demand.

What you’ll be doing:

• Working directly with customer research labs to turn unclear research goals into concrete environments, datasets, experiments and deliverables
• Designing and running post‑training loops (RL such as GRPO, preference optimization and SFT) and working out which signal actually improves the model
• Shaping and inspecting high‑fidelity trajectories and training data, with close attention to correctness, edge cases and taste
• Building reward functions, rubrics, validators and eval harnesses that hold up against reward hacking and show real capability gains
• Reading post‑training signals and experiment results across distributed GPU clusters to decide what to try next

You’ll likely be a fit if you have:

• 2+ years of experience as an applied researcher or research scientist
• Real, hands‑on RL work at a frontier lab, neolab, data lab, or on a serious independent project, with results to show for it
• A deep understanding of RL and training data, not just the tooling around it
• Experience with post‑training methods such as GRPO, preference optimization and SFT

What you won’t find here:

This role won’t suit you if you want a pure engineering execution role. The focus is research depth and customer‑facing problem solving.