View Details

About the role

Data Manager About the Role FRIDAY learns from work people actually did — recorded in the real world, not generated in simulation. And there are always more tasks worth recording than there are hours to record them. You decide what gets collected and in what proportion — which tasks, how many episodes each, across how many objects, scenes, and variations — and you revise that plan as the models show you where they are weak. You run the operation that produces it: scheduling operators across our in-house team and external partners, setting the protocol each session follows, and holding throughput against the targets the research teams plan around. You set the acceptance bar too — what counts as a usable episode, how much of each batch gets audited, and what happens to the ones that fail. Labeling is the other half of the budget. Episodes need successes and failures marked, task descriptions written, phase boundaries cut, and unusable frames excluded, and every one of those costs hours. You decide how much of it a given dataset actually needs, and work with the LBM (Large Behavior Model) team to find out which of it is earning its keep. Key Areas What You'll Do • Own the collection plan — which tasks, how many episodes, across which objects, scenes, and variations — and revise it as models reveal the gaps. • Schedule and run the operator team, in-house and external, against weekly throughput targets. • Write and maintain the session protocols operators follow, and revise them when recordings show a step is ambiguous. • Set acceptance criteria for episodes, run sampling audits, and decide what gets re-shot. • Scope labeling for each dataset — what gets annotated and to what depth — and manage the vendors or annotators who do it. • Report coverage, throughput, and reject rate so the research teams can plan against real numbers. • Close the loop with the LBM team on which data actually moved model performance, and fold that back into the plan. Required Qualifications • 3+ years running an operational process with people, targets, and a quality bar — data operations, annotation, production, logistics, or research operations. • Experience defining quality criteria and holding a team to them, including turning work away. • Able to plan against a budget of hours and money, and explain the tradeoffs behind the plan. • Comfortable working alongside engineers and reading enough about model performance to know what the data is for. Preferred Qualifications • Experience with annotation pipelines or labeling vendors. • Background in robot learning, computer vision, or another ML data domain. • Familiarity with teleoperation or motion-capture data collection. • Comfortable pulling your own numbers with SQL or a scripting language rather than waiting for them. What We Offer • Competitive compensation based on your experience and level. • Ownership of how a humanoid company spends its data budget, from the first plan onward. • Close collaboration with the research teams the data is built for. • The tools, equipment, and support you need to do your best work.