AI Engineer

🎯 The role Many of the most important questions in our work don't have established answers yet. How should an agent learn from an environment where success is hard to define? How do you know a new training signal is making models better, rather than just better at your metric? What makes an environment or task genuinely useful for learning? As a Research Scientist, you decide which of these questions are worth pursuing and design the experiments that answer them. You run those experiments hands-on. You work closely with Research Engineers who build and scale the systems, so your ideas reach production quickly. You'll help shape a research agenda measured in quarters and years, not sprints. This is a hands-on individual-contributor role. It is not a pure academic role. The measure of success is research that changes what our systems and our partners' models can do. 🚀 What you'll work on • Define research directions in post-training, reinforcement learning and learning from interaction, where there's often no obvious approach. • Design training and post-training methods, reward and grading signals, and feedback loops for agents acting in complex software environments. • Develop evaluation methods that measure real capability and catch failure modes, especially where there is no clean ground truth. • Study what makes tasks, environments and training data valuable for learning, and turn the findings into better data and environment design. • Design experiments with rigorous baselines and ablations, run them yourself, and interpret noisy or surprising results honestly. • Turn open questions into concrete, sequenced research plans with clear success criteria, including a six-month roadmap for a major area. • Work with Research Engineers to take successful methods from prototype to production. • Raise the quality of research thinking across the team through review, mentoring and technical direction. 🚀 What you'll work on • Define research directions in post-training, reinforcement learning and learning from interaction, where there's often no obvious approach. • Design training and post-training methods, reward and grading signals, and feedback loops for agents acting in complex software environments. • Develop evaluation methods that measure real capability and catch failure modes, especially where there is no clean ground truth. • Study what makes tasks, environments and training data valuable for learning, and turn the findings into better data and environment design. • Design experiments with rigorous baselines and ablations, run them yourself, and interpret noisy or surprising results honestly. • Turn open questions into concrete, sequenced research plans with clear success criteria, including a six-month roadmap for a major area. • Work with Research Engineers to take successful methods from prototype to production. • Raise the quality of research thinking across the team through review, mentoring and technical direction. 🚀 What you'll work on • Define research directions in post-training, reinforcement learning and learning from interaction, where there's often no obvious approach. • Design training and post-training methods, reward and grading signals, and feedback loops for agents acting in complex software environments. • Develop evaluation methods that measure real capability and catch failure modes, especially where there is no clean ground truth. • Study what makes tasks, environments and training data valuable for learning, and turn the findings into better data and environment design. • Design experiments with rigorous baselines and ablations, run them yourself, and interpret noisy or surprising results honestly. • Turn open questions into concrete, sequenced research plans with clear success criteria, including a six-month roadmap for a major area. • Work with Research Engineers to take successful methods from prototype to production. • Raise the quality of research thinking across the team through review, mentoring and technical direction.