Research Scientist, RL Environments

Maxonic Inc.
India

Maxonic maintains a close and long-term relationship with our direct client. In support of their needs, we are looking for a Research Scientist, RL Environments.

Job Description:

Job Title: Research Scientist, RL Environments

Responsibilities:

RL training is only as good as the tasks, graders and environments behind it. When those have flaws, models quietly pick up habits nobody intended, and teams often find out only after release.

As our Research Scientist focused on RL environments, you will own this area end to end. You’ll use a proprietary tool agent to find hidden problems in training tasks, graders and trained models. You'll prove which findings are real and turn them into fixes that AI labs, RL data vendors and training platforms can act on. You'll also shape what the agent learns to do next.

Job function

  • Check task sets before training. Run the agent over tasks, reference answers and graders to find broken tasks, false premises, graders that can be gamed, repeated templates and coverage gaps. Recommend what to fix, filter or rebalance.
  • Compare models before and after training. Show what changed besides the score, such as new shortcuts, reasoning aimed at the grader, wasted steps or odd habits from a stray reward. Trace each change back to the tasks or rewards that likely caused it.
  • Audit real-world traces. Review traces from deployed agents and evaluations to surface behaviors teams didn't expect.
  • Make findings hold up. Treat every hypothesis as unproven until it passes a held-out test with an independent checker. Write clear reports that cover what we found, how often it happens, examples, how confident we are and what to change.
  • Work directly with customers. Partner with researchers and engineers at labs, data vendors and RL as a service platforms. Learn their environment formats, run the agent on their systems when data can't leave, and walk them through the results.
  • Improve the agent. Turn what you learn on each engagement into new checks, better tools and faster setup on new environment formats.
  • Publish. Write public studies on open models and open environments that show the field what's possible.

Qualifications

  • Hands-on experience training models with reinforcement learning, such as RLHF, RL with verifiable rewards or agent RL. This includes building or debugging environments, tasks or reward functions.
  • First-hand experience of a training run that looked like a success for the wrong reasons, such as reward hacking, broken tasks or grader bugs, and tracking down why.
  • Rigorous experimental habits. You use held-out checks and honest uncertainty, and you keep what's proven separate from what's suspected.
  • Strong Python skills and comfort with the modern training and inference stack, such as PyTorch, vLLM and common RL libraries.
  • Comfort working with AI coding agents every day, and good judgment about when their output is right.
  • Clear writing and explaining. You can present the same finding to a research scientist and to an operations lead in terms each will act on.
  • Enjoyment of working directly with customers and of the ambiguity of an early-stage company.

Nice to have

  • Experience with interpretability methods such as sparse autoencoders, probing or feature analysis.
  • Experience building environments for coding, tool use, browsing, finance or other agent tasks, whether at an AI lab or a data vendor.
  • Familiarity with open environment hubs, environment formats and evaluation harnesses.
  • Exposure to topological data analysis or to clustering model behavior.
  • Public work on RL side effects, data quality or evaluation, such as papers, blog posts or open-source projects.
  • Experience doing technical work inside a customer's secure environment.

About Maxonic:

Since 2002 Maxonic has been at the forefront of connecting candidate strengths to client challenges. Our award winning, dedicated team of recruiting professionals are specialized by technology, are great listeners, and will seek to find a position that meets the long-term career needs of our candidates. We take pride in the over 10,000 candidates that we have placed, and the repeat business that we earn from our satisfied clients.

Interested in Applying?

Please apply with your most current resume. Feel free to contact Rashmi (rashmi@maxonic.com) for more details.

Score my resume against this job, free →

Questions about this job

How do I apply for the Research Scientist, RL Environments job at Maxonic Inc.?

Use the apply link on this page: you apply directly with Maxonic Inc. on LinkedIn. Before you do, score your resume against this job description on Fitboard. It shows which skills and keywords the job asks for that your resume doesn't show, so you can fix them first.

What skills does Maxonic Inc. want for this Research Scientist, RL Environments role?

The job description asks for reinforcement learning, RLHF, RL with verifiable rewards, agent RL, Python, PyTorch and vLLM. Make sure your resume shows the ones you have, in the words the job uses: applicant tracking systems look for them.

How can I get a job at Maxonic Inc.?

Apply to the Maxonic Inc. roles that match your experience, and tailor your resume to each job description: for this role that means showing reinforcement learning, RLHF, RL with verifiable rewards. A resume that matches the job's own keywords is more likely to get past screening. Fitboard scores your resume against this exact job for free.

Is this Research Scientist, RL Environments job remote or onsite?

This is based in India.

Is it free to apply? Does Fitboard charge anything?

Applying is free: you apply directly with Maxonic Inc., and this job page and its apply link cost nothing. Checking your resume against this job is free too, with 3 free scans when you sign in and no credit card needed.

Will you get shortlisted for this role? Check your match free →