SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Salary: USD 220,000 - 320,000 / annual
Replay is building production-grade reinforcement learning environments grounded in real enterprise data to serve as training datasets for frontier AI labs. The company has scaled rapidly from $0 to multi-eight-figure run rate in months, backed by top-tier investors including Floodgate and Afore.
As an RL Environment Engineer, you will design and build production-grade RL environments that transform real business workflows and de-identified enterprise data into realistic, verifiable training environments. You'll work directly with frontier lab researchers to understand capability gaps and design task suites that expose and train against them.
Key responsibilities include:
- Design and build production-grade RL environments from task specification through stable, versioned release
- Write reward functions and programmatic verifiers that score outcomes reliably and resist reward hacking
- Build high-fidelity evaluation environments with seeded failure modes and reliable model/agent harnesses
- Transform de-identified enterprise data into realistic state, tools, files, and context for agents
- Maintain reproducibility, observability, and safety in environment rollouts with clear version tracking
- Collaborate with frontier lab researchers to understand capability gaps and design appropriate task suites
- Measure task difficulty and solvability by running frontier models, then tune for usable training signals
- Ship quickly and establish standards and patterns for how Replay builds tasks and environments
This is a hands-on engineering role with high degree of ownership. You'll carry outsized responsibility in a small, fast-growing company and partner directly with the fastest and most important companies in the world.
REQUIREMENTS:
- 3-5+ years of ML/software engineering experience shipping production systems
- Personally built production-grade RL environments used in real training or evaluation (not prototypes or research demos)
- Ability to turn ambiguous researcher goals into well-scoped, working systems
- Clear grasp of environment design tradeoffs: task specification, state and tool design, determinism, difficulty calibration, solvability
- Built evaluation or agentic harnesses with comfort in evaluation rigor: precision, recall, calibration, imperfect labels, distribution shift
- High-ownership, action-oriented mindset; take problems from idea to shipped
- Startup experience with taste for broad ownership and ability to work with evolving requirements
- Clear written and verbal communication with researchers, engineers, and non-technical partners
BONUS:
- Background at or working with a frontier AI lab, or research-adjacent engineering
- Forward-deployed or customer-facing engineering experience
- Experience with sensitive enterprise data or high-trust production systems
- Founder or early employee at a startup
