Software Engineer

Acceler8 Talent; San Francisco (CA); On-site; USD 225,000 - 275,000; Full time

Benefits offered by this job

Equity

Job summary

Acceler8 Talent is hiring a Software Engineer to design reinforcement-learning environments for frontier model labs. You will turn research objectives into scalable environments, tasks, reward signals, and evaluation rubrics that drive model learning and behavior.

You’ll work on RL environments across coding, finance, and enterprise workflows, building robust pipelines and experiments that reveal why models succeed or fail, with in-person collaboration in San Francisco.

Qualifications

• Experience applying RL to LLM post-training, agents, or sequential decision-making.
• Experience building environments, reward functions, evaluation tasks, or scoring systems.
• Familiarity with RLHF, RLVR, policy optimization, or preference-based learning.
• Ability to design controlled experiments and interpret noisy results.

Responsibilities

• Turn research objectives into scalable RL environments and experiments.
• Design and implement RL environments, reward signals, and evaluation rubrics.

Skills

RL in LLM post-training

Environments & rewards

RLHF/RLVR

Experiment design & analysis

Job description

Software Engineer, Reinforcement Learning Environments

$250k base + equity + bonus

I’m working with a highly profitable AI research infrastructure company building reinforcement-learning environments for frontier model labs.

The team creates the tasks, simulations, reward signals, evaluation rubrics, and expert trajectories used to improve how advanced models reason and act. These are not generic annotation datasets. They are structured RL environments designed to expose failure modes, measure capability, and produce useful learning signals.

This role will directly influence model post-training by turning research objectives into environments and experiments that can run at scale.

You’ll work on problems such as:

• Designing RL environments for coding, finance, and enterprise workflows
• Creating tasks that expose meaningful model and agent failure modes
• Building reward functions and evaluation rubrics for RLHF and RLVR
• Analyzing trajectories to understand why agents succeed or fail
• Improving the quality and reliability of training signals
• Measuring how environment and dataset changes affect model capability
• Developing real-world and synthetic-data pipelines

Looking for engineers who have:

• Experience applying RL to LLM post-training, agents, or sequential decision-making
• Experience building environments, reward functions, evaluation tasks, or scoring systems
• Familiarity with RLHF, RLVR, policy optimization, or preference-based learning
• The ability to design controlled experiments and interpret noisy results

This is an opportunity to build the reinforcement-learning environments and reward systems that directly shape how frontier models learn, reason, and improve.

The company operates in person from San Francisco and values speed, ownership, experimental judgment, and measurable results.