Founding Engineer – RL Environments & Data

CGS; San Francisco (CA); On-site; USD 150,000 - 230,000; Full time

Benefits offered by this job

Unlimited AI spend

Daily lunch provided

Equity—meaningful

Job summary

CGS in San Francisco is seeking a Founding Engineer focused on RL Environments & Data to turn real-world agent failures into simulated environments and evaluation benchmarks. You will build simulated business software environments and tools to measure task completion, connecting production performance with model training.

This is a founder-level, hands-on role with two founding engineering hires; ownership of RL environments, datasets, and evaluation strategy; deep collaboration with founders;

Qualifications

• 1+ years of relevant experience.
• Concrete example of building or contributing to an RL environment, benchmark, dataset or model evaluation system.
• Experience with synthetic data, post-training or model evaluation.
• Ability to simulate business software environments that can withstand complex agent tool interactions.
• Experience designing graders and evaluations robust against reward hacking.
• Strong programming and problem-solving skills.
• Willingness to work full-time in person in San Francisco, Monday to Friday.

Responsibilities

• Analyse real agent interaction traces to identify meaningful model failures.
• Build realistic simulated business environments including email inboxes, CRMs, databases, records and file systems.
• Develop reinforcement learning environments, benchmarks and evaluation systems.
• Design robust graders that accurately measure task completion and resist reward hacking.
• Create synthetic datasets and support post-training workflows.
• Build evaluations whose rankings correlate with real-world production performance.
• Identify weaknesses in existing benchmarks and develop better measurement methods.
• Turn research and production insights into environments and datasets for AI labs.

Skills

RL environments

Programming

Problem-solving

Model evaluation

Synthetic data

Datasets

Job description

Founding Engineer – RL Environments & Data

Working model: Full-time, in-office (5 days a week)

Working hours: 9 AM – 5 PM, Monday to Friday

Experience: 1+ years (strong practical experience matters more than tenure)

About the company

A profitable, seed-stage AI company in San Francisco is building autonomous agents capable of completing entire business tasks. Its agents already handle workflows such as accounts receivable through Slack and scheduling through email. The company generates hundreds of thousands of agent interaction traces every week, with customers acquired primarily through referrals.

The team is now expanding with two founding engineering hires to build the environments, datasets and evaluation systems that make its agents more capable and reliable.

The role

As a Founding Engineer focused on RL Environments & Data, you'll turn real-world agent failures into realistic simulated environments, robust graders and meaningful benchmarks. You'll build simulated business software environments, including inboxes, CRMs, records and file systems, where agents can interact with tools and complete complex tasks.

Your work will connect real production performance with model training and evaluation. You'll help identify which failures matter, create datasets and environments to address them, and develop benchmarks that accurately reflect how agents perform in the real world. This is an engineering and data-focused role rather than traditional product engineering or purely academic research. You'll work directly with the founders and have significant ownership over the technical direction.

What you'll do

• Analyse real agent interaction traces to identify meaningful model failures.
• Build realistic simulated business environments, including email inboxes, CRMs, databases, records and file systems.
• Develop reinforcement learning environments, benchmarks and evaluation systems.
• Design robust graders that accurately measure task completion and resist reward hacking.
• Create synthetic datasets and support post-training workflows.
• Build evaluations whose rankings correlate with real-world production performance.
• Identify weaknesses in existing benchmarks and develop better ways to measure agent capabilities.
• Turn research and production insights into environments and datasets that can be used by AI labs.

What we're looking for

• 1+ years of relevant experience, with demonstrated practical ability.
• A concrete example of building or contributing to an RL environment, benchmark, dataset or model evaluation system used to train or evaluate models.
• Experience with synthetic data, post-training or model evaluation. Human annotation or expert labelling alone is not sufficient.
• Ability to simulate business software environments that can withstand complex agent tool interactions.
• Experience designing graders and evaluations that are robust against reward hacking.
• Strong programming and problem-solving skills.
• Willingness to work full-time in person in San Francisco, Monday to Friday.

What makes you stand out

• Experience building environments or evaluation systems used by frontier AI labs.
• Experience identifying and fixing reward hacking or flaws in widely used benchmarks.
• Background in synthetic data generation, RLHF or model post-training.
• Experience building agent benchmarks and simulated business workflows.
• Previous founder experience or participation in a founder fellowship.
• Strong achievements in mathematics, programming, debate or competitive games.
• A clear understanding of which model failures and datasets have meaningful commercial value.
• Join a profitable, seed-stage AI company as one of just two founding engineering hires.
• Own a critical part of the company's RL environments, data and evaluation strategy.
• Work directly with the founders and influence the technical direction.
• Work on real-world agent traces generated at significant scale.
• Contribute to data and RL contracts with leading AI labs.
• No time tracking: the focus is on output and impact.
• Unlimited AI spend and daily lunch provided.
• Meaningful equity and the opportunity to make an early impact.
• Founder screen (30 minutes): Background, relevant experience, communication, motivation and collaboration fit.
• Technical interview (60 minutes): Deep dive into a project from your CV, covering design decisions, personal ownership, environments, graders and software simulations.
• Practical assessment: Either a 2–4 day work trial or a 1–2 day take-home assignment.
• Reference checks and offer.