Agentic RLE Author – AI Safety Evaluation

Remote · Freelance / Contract · Paid per task · Immediate start
**
**About the role

We're hiring RLE (Reinforcement Learning Environment) Authors.
This is not prompt writing or annotation. You'll build the evaluation environments that test whether AI agents behave safely during real coding tasks.
SolumnAI works with frontier AI teams on AI safety, security and alignment. Your environments will be used to measure and improve how advanced coding agents handle security-sensitive work.

What you'll do

• Design complete, Dockerized scenario packages from scratch
• Write clear task instructions and build realistic repository fixtures
• Write working reference solutions in Python
• Build deterministic verifiers and test harnesses that grade agent behavior reliably
• Review agent trajectories to find where an agent went wrong and why
• Revise your packages based on quality-review feedback

What we're looking for

• Strong Python: you'll write real code (reference solutions, test harnesses, containerized environments), not pseudocode
• Security awareness: you understand how vulnerabilities work, and the difference between a proper fix and a superficial patch
• Docker fluency: building and validating self-contained environments feels routine to you
• Trajectory review: you can read an agent's reasoning trace and diagnose failures
• The ability to think adversarially and constructively at the same time Strong plus
• Experience with RLHF, AI evaluation, benchmark authoring or red-teaming

How pay works

• You're paid per task, not per hour
• Each task is paid once it passes our quality review
• Rate: [₹X – ₹Y per accepted task, depending on complexity]
• Steady task volume for strong authors, with flexible hours

Engagement details

• 100% remote
• Freelance / contract
• Immediate start
• Work on your own schedule, within task deadlines How to apply Apply with your resume and, if you have them, links to GitHub or to past evaluation, security or benchmark work. Shortlisted candidates will need to complete an assessment.