Reinforcement Learning Engineer

This is a fully remote position, open to applicants in United States.

📋 Description

• Design, construct, and implement high-throughput RLVR environments that assess LLM action sequences against deterministic execution results.

• Create automated testing frameworks, sandboxes, and verification engines that transform vulnerability research into binary pass/fail reward metrics.

• Integrate Bugcrowd’s Mayhem automated analysis platform along with real-world vulnerability feeds into scalable RL environment generation processes.

• Architect safe, isolated, and reproducible execution environments utilizing Docker, BuildKit, or Nix.

• Collaborate with researchers from Anthropic, OpenAI, and Cohere to establish benchmark formats, observation spaces, and verifiable evaluation metrics.

• Implement telemetry, ground-truth verification algorithms, and trajectory logging to examine agent reasoning and prevent reward manipulation or false positives.

• Develop low-level instrumentation and debugging tools for monitoring memory states, process executions, and network activities.

• Enhance infrastructure performance and environment reset latency for large-scale parallel sampling and distributed RL training.

• Evaluate cutting-edge AI models across offensive and defensive security challenges, including automated fuzzing, exploit payload generation, and patch validation.

⛳️ Requirements

• Understanding of RL training workflows employed by contemporary LLM systems, particularly execution-based feedback or Reinforcement Learning from Verifiable Rewards (RLVR).

• Proficient in developing applications using Python.

• Competence in low-level systems programming with C.

• Experience with Rust is highly desirable.

• Strong grasp of software vulnerabilities, binary exploitation, fuzzing methodologies, or program analysis.

• Familiarity with DevOps pipelines, such as GitHub Actions.

• Experience in creating reproducible builds utilizing Docker, BuildKit, or Nix.

• Comfortable working with Linux systems and performing low-level debugging.

• Experience in developing or working with benchmark environments such as CTFs, SWE-bench, security challenges, or execution sandboxes.

• Preferred: experience in designing custom reward functions, ground-truth verifiers, or automated grading systems for AI safety and reasoning models.

• Preferred: background in low-level program analysis tools, sanitizers like ASan/MSan, compiler instrumentation, or automated exploit generation tools.

• Preferred: proven history of participating in or creating competitive cybersecurity benchmarks, CTFs, or open-source AI evaluation frameworks.

• Capability to fulfill all physical demands with or without reasonable accommodation.

• Ability to remain stationary for 50% of the time.

• Capacity to carry or relocate a laptop as necessary throughout the workday.

• Integrity to uphold confidential information with the utmost discretion.

• Successful completion of relevant background checks.

🏝️ Benefits

• Eligibility for a discretionary bonus program or commission plan.

• Reasonable accommodations for disabilities.

• Flexible work-from-home arrangement.