About the role
The reinforcement learning team builds the environments, reward signals and training infrastructure that teach models to act over long horizons. A model that writes a good paragraph is not the same as one that completes a forty-step task, and closing that gap needs verifiable tasks, stable policy optimisation and a rollout system that does not waste accelerators.
Clexerl Go is the proving ground. Its agents schedule, triage and coordinate on behalf of people, and each real failure becomes a candidate environment. Research engineers here own the loop from task design to trained policy, working beside the Reasoning and Agents teams, who supply tasks and consume the resulting checkpoints.
The team treats reward design as an adversarial problem. Every environment ships with a written note on how it might be gamed, and someone is assigned to try. Throughput matters as much as ideas, so engineering and research work on the same board.
What you’ll do
- Build RL environments with programmatic, verifiable rewards for coding, tool use, planning and multi-turn interaction.
- Implement and scale policy-optimisation algorithms such as PPO and GRPO-style methods for large language models.
- Engineer asynchronous rollout and learner pipelines that keep inference and training accelerators busy without staleness harming learning.
- Hunt for reward hacking by auditing trajectories, writing adversarial tests and tightening verifiers.
- Monitor training with the right diagnostics, including entropy, KL, advantage statistics and per-task success curves.
- Partner with researchers to turn a promising idea into a stable, reproducible training run.
What you’ll bring
- Strong engineering background in Python and a deep learning framework, with experience building systems that run reliably for days.
- Working knowledge of reinforcement learning: policy gradients, advantage estimation, off-policy corrections and exploration.
- Experience with large language model inference engines, including batching, KV-cache management and sampling.
- Skill in designing evaluations or environments where the metric genuinely reflects the goal.
- Comfort with distributed systems concepts such as queues, backpressure, idempotency and failure recovery.
- A habit of suspecting your own results, especially when the reward curve looks too good.
Nice to have
- Experience with sandboxed code execution, container orchestration or secure tool environments.
- Prior research on process rewards, credit assignment or long-horizon exploration.
- Background in multi-agent or self-play training.
- Experience with speculative decoding or other inference optimisations in a training loop.
Work you might take on
- 01
Build a sandboxed environment suite for multi-step software tasks with unit-test rewards and a documented list of known exploits.
- 02
Replace a synchronous rollout loop with an asynchronous one and quantify the throughput gain against policy staleness.
- 03
Create a calendar and scheduling environment modelled on real Clexerl Go failure cases, with partial-credit rewards.
- 04
Implement a trajectory auditor that flags reward hacking automatically and routes suspicious runs to a human reviewer.
Compensation
$320K – $520K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.