RL Engineer

Reinforcement Learning Engineer

Location: San Francisco, CA — on-site, five days per week

Employment: Full-time

Compensation: $220,000–$290,000 base + $200,000–$300,000 bonus + equity

About the company

Our client builds the training data, reinforcement learning environments and evaluation infrastructure used by leading AI labs to improve foundation models.

The team works directly with researchers to identify where models struggle and turn real-world problems into training tasks, reward signals and evaluations. Its work spans software engineering, finance and complex enterprise workflows.

This is an opportunity to join a growing AI company where your work directly influences how models learn, reason and complete useful tasks.

The role

As a Reinforcement Learning Engineer, you will design and build environments that challenge AI agents to complete realistic tasks. You will define what successful behaviour looks like, implement reliable ways to measure it, and investigate whether improvements reflect genuine capability gains.

The role combines hands-on software engineering with applied research. You will move quickly from a hypothesis to a working experiment, developing tasks, datasets and evaluation systems that feed into model training at scale.

We are looking for either an engineer with direct experience building RL environments or a hands-on RL/ML researcher with strong intuition for agent task design and the engineering ability to turn ideas into working systems.

What you’ll do

• Build RL environments and agent tasks: Develop realistic, executable tasks across domains such as coding, finance and enterprise workflows.
• Design rewards and evaluations: Create and refine scoring rubrics, verifiers and reward signals for reinforcement learning from human feedback (RLHF) and verifiable rewards (RLVR).
• Investigate model failures: Identify where agents struggle, diagnose misleading evaluation results and address opportunities for reward exploitation.
• Develop training data: Build and maintain real-world and synthetic data pipelines, selecting tasks and data slices that expose meaningful capability gaps.
• Run experiments: Test environment designs, data collection strategies and reward formulations, using results to guide the next iteration.
• Measure data quality and impact: Develop quantitative approaches to assessing dataset quality, diversity and downstream effects on model capability and alignment.
• Improve human feedback workflows: Analyse annotator behaviour and experiment with ways to improve the quality and consistency of collected data.
• Collaborate with researchers: Translate partner labs’ training objectives into concrete environment, data and evaluation specifications.
• Build reliable infrastructure: Use containerisation and strong engineering practices to make environments and experiments reproducible and scalable.

What we’re looking for

• Approximately 1–8 years of relevant software engineering or machine learning experience.
• Direct experience building reinforcement learning environments, or a strong RL/ML research background with demonstrated ability to design meaningful agent tasks.
• Strong backend or full-stack engineering skills, with depth in Python, TypeScript or another relevant programming language.
• Familiarity with reinforcement learning methods, particularly their application to LLM post-training.
• The ability to design experiments, interpret imperfect results and turn findings into practical improvements.
• Evidence of technical excellence through relevant industry work, high-quality AI research or substantial technical projects.
• A strong academic foundation in computer science or a closely related technical discipline.
• A practical approach to execution, including the willingness to own detailed implementation, debugging and data-quality work.

Particularly relevant experience

• RL environment development, including internships or substantial research projects.
• Agent benchmarking, AI safety evaluations or reward-hacking research.
• Building automated verifiers, evaluation harnesses or reward models.
• Designing synthetic tasks or measuring training-data quality and diversity.
• Docker or similar containerisation tools.
• Published AI research or meaningful open-source contributions.
• Early-stage startup experience, including founding or early engineering roles.

Why join?

• Work directly with research teams at leading AI labs.
• Build environments and evaluations that influence model training at scale.
• Own projects from initial idea through implementation and experimentation.
• Combine research problems with substantial software engineering ownership.
• Receive a competitive base salary, bonus package and equity.

Location and work authorisation

The team works together in its San Francisco Financial District office, five days per week.

Support is available for certain work-authorisation routes, including H-1B transfers, O-1, TN and STEM OPT, subject to individual circumstances.

Interested?

Contact Thuvaa Baskaran at Surge Group with your CV or LinkedIn profile. Links to relevant research, repositories or projects are welcome, particularly examples of environments, agent tasks, reward systems or evaluations you have personally built.