Applied Researcher – Continual Learning

This is a fully remote position, open to applicants in United States, +2 more locations.

📋 Description

• Develop ongoing learning systems that transform agent trajectories, corrections, and failures into new tasks and updates for models.

• Analyze streaming reinforcement learning, batch post-training, and distillation methods.

• Identify which enhancements should be integrated into model weights, persistent memory, skills, prompts, and subagent states.

• Monitor the acquisition of skills, transfer, forgetting, and the decline of plasticity across extended sequences of tasks.

• Convert workflows into tasks, environments, evaluations, and reward signals.

• Conduct post-training experiments against real-world baselines and implement effective strategies.

• Manage update frequency, policy delay, task drift, and exploration strategies.

• Assess checkpoints based on previous skills, unfamiliar tasks, and overall capabilities.

• Create reproducible task generation and training pipelines utilizing version-controlled data and harnessing tools.

• Choose production traces and interventions, transform them into training signals, and establish regression gates.

• Develop evaluations for retained learning, transfer, regression, and rollback mechanisms.

• Take responsibility for methods related to selective consolidation, replay, adaptation, memory, and both online and offline reinforcement learning.

• Lead field experiments, benchmarks, and open research initiatives focused on continuous improvement.

• Collaborate directly with external partners, customers, users, or research associates.

⛳️ Requirements

• Relevant background in continual or online learning, reinforcement learning, meta-learning, memory systems, post-training data loops, adaptive production machine learning, or evaluation in varying distributions.

• Practical experience in constructing and troubleshooting model training, post-training, agent evaluation, or closely related experimental processes.

• Strong experimental judgment, including meaningful baselines, thorough trace examinations, clear failure attribution, and a readiness to adjust direction based on evolving evidence.

• Proficiency in Python and the capability to create artifacts that can be executed, inspected, and extended by other researchers and engineers.

• Comfort in handling ambiguous challenges and keeping partners and team members informed as solutions are developed.

🏝️ Benefits

• Equity incentives

• Flexible work arrangements (San Francisco, New York, London, or remote)

• Visa sponsorship and relocation assistance

• Professional development budget

• Team off-sites and conference participation