Applied Researcher – Computer Use

This is a fully remote position, open to applicants in California.

📋 Description

• Develop systems for post-training and evaluation of multimodal agents that operate across browsers, desktops, and various application workflows.

• Transform actual workflows and traces into environments, rewards, and experiments that facilitate consistent improvement.

• Create long-term computer-use environments based on real workflows while avoiding shortcuts, leakage, and unsafe side effects.

• Train agents to adapt to interface changes, handle partial failures, and manage ambiguous intermediate states.

• Formulate evaluations that differentiate true completion from behavior that merely appears plausible, focusing on correctness, efficiency, permissions, and appropriate escalation.

• Construct multimodal environments derived from partner workflows, demonstrations, tools, interface states, and outcome criteria.

• Develop agents for browsers, desktops, and cross-application use with reliable sandboxing and boundary enforcement.

• Conduct training and evaluations using demonstrations, interventions, production failures, synthetic tasks, and verifiable outcomes.

• Execute experiments on recovery and generalization across applications, layouts, permissions, and deployment contexts.

• Collaborate directly with customers, users, or research partners while maintaining a robust research agenda.

• Transform successful reusable work into shared research, open-source systems, and product capabilities.

⛳️ Requirements

• Relevant experience in multimodal agents, computer usage, browser or desktop automation, reinforcement learning (RL), imitation learning, human-computer interaction (HCI), embodied agents, or long-horizon agent evaluation.

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

• Strong experimental judgment, including the ability to establish meaningful baselines, conduct thorough trace inspections, accurately attribute failures, and adapt direction based on new evidence.

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

• Comfort in managing ambiguous challenges and keeping partners and teammates informed as you work towards resolutions.

• Publications, open-source contributions, benchmarks, or delivered products regarded as indicators of outstanding work.

🏝️ 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