Founding Engineer, RL Environments + Data

We’re supporting a client with the recruitment requirement outlined below. If your recruiting expertise and networks align with the role, please apply to express your interest.

Once approved to work on the opportunity, you’ll receive the full brief, including the client name, role details, and commercial terms.

Role Details

Location: San Francisco, CA

Industry: AI Software

Salary: $150,000–$300,000

Employment Type: Full-time

Work Arrangement: In-office, Monday–Friday

Experience: 1–7 years

Visa / Sponsorship: Available

Team Size: 1–10

About the Company

An early-stage AI software company developing a platform that enables AI agents to perform complete business workflows, such as accounts receivable and scheduling, through tools like Slack and email.

The company has raised approximately $6M in seed funding, generates around $4.5M in annual revenue, and is profitable. It produces hundreds of thousands of agent traces each week, using them to identify model failure modes, improve its platform, and develop evaluation data and research for AI labs.

About the Role

The company is looking for a founding engineer specializing in reinforcement learning environments, synthetic data, and model evaluation.

The main objective is to turn large volumes of real agent interaction traces into reliable environments, benchmarks, and datasets that can evaluate and improve AI agents.

A key priority is building simulated business environments—including inboxes, CRMs, records, and files—that reproduce realistic workflows and test how agents perform across extended tasks. The benchmarks must provide meaningful results that correlate with real-world production performance.

This is an environments and data-focused engineering role, rather than a product-facing full-stack position or pure research role.

Responsibilities

• Build synthetic environments that simulate the software and workflows agents use to complete business tasks.

• Develop reliable graders and evaluation systems resistant to reward hacking.

• Create benchmarks that accurately reflect agent performance in production.

• Analyze agent traces to identify meaningful and previously unquantified model failure modes.

• Convert identified failures into environments, datasets, and evaluation tools.

• Support data and reinforcement learning contracts with AI labs.

• Improve the company's evaluation infrastructure and agent performance.

Requirements

• Experience with RL environments, synthetic data, post-training, benchmark development, dataset building, or model evaluation.

• Hands-on experience building or contributing to environments, benchmarks, or datasets used for model training or evaluation.

• Ability to identify meaningful model failures and distinguish valuable data from noisy traces.

• Understanding of evaluation design, graders, reward hacking, and model failure modes.

• Ability to simulate software systems such as inboxes and CRMs that can withstand agent tool calls.

• Strong technical ownership and ability to explain previous projects and design decisions in depth.

• Commercial understanding of how datasets, environments, and evaluations can create value.

• Important: Human-expert or annotation experience alone is not sufficient. Candidates should have experience building data, environments, benchmarks, or post-training systems.

Nice to Have

• Experience discovering substantive flaws in widely used benchmarks.

• Experience identifying and fixing reward hacking in model graders.

• Environments or benchmarks adopted by frontier AI labs for training or evaluation.

• Founder background or participation in YC, SPC, Thiel Fellowship, or similar programs.

• Strong competitive achievements in mathematics, programming, debate, or games.

• Understanding of the commercial value of benchmarks and datasets for model evaluation.

Benefits

• Medical, dental, and vision insurance.

• Daily lunch.

• 401(k).

• Unlimited PTO.

• Unlimited AI spend.

Interview Process

• Founder Screen

• Technical Interview

• Work Trial or Take-Home Assessment (the format depends on candidate availability)

• Reference Checks.

• Offer.