Senior Environment Engineer, Long-Horizon Tasks

Reinforced Learning; Town of Tonawanda (NY); On-site; USD 140,000 - 210,000; Full time

Job summary

Reinforced Learning seeks an experienced engineer to own and scale complex, long-horizon environments for training and evaluation. You will design environments where state persists across many tool calls and events, coordinating interfaces between environments, verifiers and the training stacks used by customers.

You will lead cross-team design reviews, push for reproducibility, and research where agents fail as horizon length grows, shaping robust, evolving simulation platforms for real-world

Qualifications

• Five or more years building distributed or simulation systems in production.
• Deep comfort with concurrency, state machines and failure modes.
• You've led technical design across a team and can explain a tradeoff in writing.
• Interest in how agents fail and what a world needs to contain to surface it.

Responsibilities

• Environments where state persists across many tool calls and the right answer depends on what happened earlier.
• Event schedules, background processes and interruptions that make a world behave like a workplace.
• The interfaces between environments, verifiers and the training stacks our customers run.
• Design review across the team, raising the bar on reproducibility and reward integrity.
• Measurement, with research, of where agents break as horizon length grows.

Skills

Distributed systems

Simulation systems

Concurrency

State machines

Failure modes

Technical leadership

Written communication

Tools

Game engines

Discrete event simulation

Job description

The role

Most environments test a few steps. Ours increasingly need hundreds: tasks that span days of simulated time, several systems and partial information. You'll own the hardest environments we build and set the patterns the rest of the team follows.

What you'll own

• Environments where state persists across many tool calls and the right answer depends on what happened earlier.
• Event schedules, background processes and interruptions that make a world behave like a workplace.
• The interfaces between environments, verifiers and the training stacks our customers run.
• Design review across the team, raising the bar on reproducibility and reward integrity.
• Measurement, with research, of where agents break as horizon length grows.

What you bring

• Five or more years building distributed or simulation systems in production.
• Deep comfort with concurrency, state machines and failure modes.
• You've led technical design across a team and can explain a tradeoff in writing.
• Interest in how agents fail and what a world needs to contain to surface it.

Helpful, not required

• Game engines, discrete event simulation or workflow engines.
• Hands-on RL post-training of language models.