Remote job
Research Engineer, Frontier Data
Job details
About this role
Role overview
Design and deliver datasets, reinforcement learning environments, and evaluations that help improve advanced AI models. You’ll work with research and engineering teams to turn evolving training goals into clear technical plans, then build quality checks and verification methods for bespoke projects. The work may focus on coding agents, reinforcement learning, multimodal or STEM reasoning, or embodied AI systems.
Responsibilities
- Translate research goals into data and environment specifications, including target skills, task difficulty, coverage, failure cases, and success measures. - Assess technical feasibility and define assumptions, acceptance criteria, and evaluation needs. - Create task suites, ground-truth signals such as tests or verifiers, and interfaces such as APIs, tool schemas, or state representations. - Write detailed quality rubrics and use audits to identify errors, ambiguity, data leakage, reward exploitation, and distribution shifts. - Analyze recurring failures and turn findings into concrete guidance that improves data production and quality. - Work with research and production teams to balance delivery speed with correctness, realism, diversity, and measurable model improvement.
Requirements
- Experience building or improving deep learning systems, particularly where data curation, denoising, synthetic data, or rigorous evaluation is important. - Ability to translate ambiguous technical goals into specific requirements, tests, and acceptance criteria. - Strong analytical judgment and attention to detail when reviewing data, tasks, and model evaluations. - Experience in at least one relevant area: software engineering agents, reinforcement learning, multimodal data, STEM reasoning, or embodied AI. - Ability to communicate findings clearly and use evidence from failures to guide iterative improvements.
Benefits and work setup
- Remote work from Brazil or Colombia. - Direct collaboration with an AI research team, with the opportunity to contribute to advanced model training and evaluation work.