Applied Research - Continual Learning

Role overview Build systems that help agents improve as they encounter new tasks, using their trajectories, corrections, and failures as learning signals. You’ll compare online and batch learning approaches, distillation, and persistent agent state, then measure whether skills transfer over time without eroding earlier capabilities.

Responsibilities - Develop methods for selective learning, replay, adaptation, memory, and online or offline reinforcement learning. - Turn useful production traces and interventions into training examples, with regression checks before updates reach deployment. - Compare which improvements belong in model weights versus memory, tools, prompts, data, or harness state. - Evaluate skill acquisition and transfer alongside forgetting, regressions, and changes in general capability. - Build reproducible task-generation and training pipelines with versioned data and harnesses, enabling changes to be traced and rolled back. - Work with external collaborators to run field experiments and develop benchmarks and research on measurable continual improvement.

Requirements - Experience in continual or online learning, reinforcement learning, meta-learning, memory systems, post-training data loops, adaptive production ML, or evaluation under changing distributions. - Hands-on experience building and debugging model training, post-training, agent, evaluation, or similar experimental workflows. - Strong experimental judgment, including meaningful baselines, careful trace inspection, and clear failure attribution. - Strong Python skills and ability to produce artifacts other researchers and engineers can run, inspect, and extend. - Comfort owning ambiguous problems and keeping collaborators informed as work progresses. - Evidence of substantial work, such as publications, open-source systems, benchmarks, or shipped products; be ready to describe a difficult project and how evidence changed your approach.

Benefits and work setup - Full-time, with work locations listed in San Francisco, New York, London, or remotely. - Listed cash compensation is $150,000–$300,000, plus equity incentives. - Visa sponsorship and relocation support, a professional development budget, and team off-sites and conference attendance are offered.