Research Engineer

Role overview Help improve advanced AI systems by designing high-quality datasets, reinforcement learning environments, and evaluations. This research-facing engineering role turns model-development goals into concrete tasks and quality standards, then uses careful audits and evidence from model behavior to refine the resulting data and environments.

Responsibilities - Translate research objectives into specifications covering target capabilities, failure cases, difficulty, coverage, and success measures. - Create rubrics, counterexamples, and boundary cases that define acceptable data and task behavior. - Audit datasets and environments for subtle errors, ambiguity, information leakage, reward exploitation, inconsistent assumptions, and distribution shifts. - Use error categories, slice-based metrics, and model behavior to guide iterative quality improvements. - Design task suites, ground-truth signals such as verifiers or tests, and environment interfaces such as APIs, tool schemas, and state models. - Contribute in one or more areas, including coding agents, reinforcement learning, multimodal reasoning, STEM, or vision-language agents.

Requirements - Roughly four to five years of experience building or improving deep learning systems. - Strong experience with data quality, curation, denoising, synthetic data, or rigorous evaluation. - Ability to turn ambiguous research goals into precise specifications and measurable quality criteria. - Careful analytical skills for finding errors and assessing task correctness, realism, diversity, and difficulty. - Systems thinking: ability to model application interfaces, data structures, and realistic workflow constraints. - Relevant depth in at least one of the role’s capability areas, such as software engineering agents, verifiable STEM tasks, multimodal data, or RL environments.

Benefits and work setup Remote role open to candidates based anywhere in Colombia. The work offers direct collaboration with AI research and engineering teams, high autonomy, and rapid iteration.