> Markdown version of [/jobs/ext/4413655-research-engineer-scientist-reinforcement-learning](https://www.wearedevelopers.com/jobs/ext/4413655-research-engineer-scientist-reinforcement-learning). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Research Engineer / Scientist - Reinforcement Learning

- **Company:** Prime Inc.
- **Location:** San Francisco, CA, United States (Remote available)
- **Salary:** $150,000.0 - $350,000.0
- **Contract:** Permanent contract
- **Skills:** Training Data, Computer Programming, Software Debugging, Distributed Computing Environment, Python (Programming Language), Reinforcement Learning, Pytorch, Free and Open-Source Software, Machine Learning Operations
- **Published:** October 15, 2026
- **Apply:** https://www.indeed.com/viewjob?jk=93cf7b5c1d8b6c19

## About the Role

* Demonstrated depth in at least one of RL algorithms, empirical research on model training, or distributed ML systems. This could come through research, open-source contributions, or systems you've built.
 * Strong programming skills in Python and experience with PyTorch or a comparable ML framework.
 * Ability to design controlled experiments, interpret results, and debug unexpected model or system behavior.
 * Experience taking a research idea or systems change through implementation, testing, and evaluation, with a clear account of what worked and what didn't.

Nice to have

 * Experience with distributed training frameworks or asynchronous rollout pipelines.
 * Experience diagnosing unstable RL runs, including policy-staleness or numerical issues.
 * Work on coding agents, verifiable rewards, or long-horizon environments.

## Description

Algorithm research

 * Develop and test RL methods for reasoning, coding, and long-horizon tasks, including changes to training objectives and policy optimization.
 * Investigate stability in asynchronous RL, including policy staleness, trainer/sampler mismatch, and numerical differences between training and rollouts.
 * Work with data and evaluation researchers to understand how task selection, rewards, and training data affect what the model learns.

Scaling science

 * Design experiments to understand how RL performance and learning dynamics change with model size, training compute, and task horizon.
 * Use controlled ablations and evaluations to distinguish learning gains from reward exploitation, evaluation artifacts, or training bugs.
 * Turn experimental findings into training recipes, then test whether the improvements hold across larger models and longer runs.

RL frameworks

 * Extend prime-rl to support new algorithms, models, and asynchronous training workflows.
 * Improve the performance and numerical correctness of research runs across trainers, rollout workers, and environments, working with the infrastructure team.
 * Build diagnostics that make unstable runs and performance regressions easier to reproduce, and help distinguish algorithm problems from implementation or runtime issues.