The company is building AI-powered humanoid robots, with its flagship robot Apollo designed to work alongside people. The first target industries are manufacturing and logistics, with healthcare and home use planned later.
The team is bringing Apollo to market at scale and working on hard problems such as safety, commercialization, and mass production.
This is a Staff or Principal Technical Program Manager role that leads the programs training Apollo's AI. The role covers the whole physical AI training lifecycle: compute infrastructure, the data engine, pretraining, mid-training, and post-training runs, and the route from finished models onto robots in the field.
You will work as a technical peer to research and infrastructure leaders rather than as a scheduler. You will set how training programs are planned and resourced, run the build-out and utilization of cloud and on-premises GPU capacity, and convert research goals into executable plans with defined gates and decisions. As the function grows, you will set the program management practices the team uses to train and ship models.
Responsibilities
Own the strategy and delivery of training compute across cloud and on-premises GPUs, including capacity forecasting, build-versus-buy decisions, vendor and cloud partnerships, cluster bring-up, and how compute is used, costed, and allocated across research teams.
Run pretraining, mid-training, fine-tuning, and distillation as structured programs. Set compute budgets, data freeze dates, checkpoint and evaluation gates, and go/no-go criteria, and lead the response when a run diverges or infrastructure fails.
Own the data engine that turns model needs into training data from teleoperation, human demonstration, and simulation. Turn research requirements into collection plans, quality standards, and SOPs, and grow data throughput through the pipelines and storage behind it.
Set up a rigorous evaluation and release process covering simulation and real robots. Build statistically sound benchmarks, regression gates, and clear promotion criteria so release decisions are based on evidence rather than demos.
Own the path from trained checkpoint to on-robot inference, including model optimization for edge hardware and integration with software, controls, and firmware. Build the loop that feeds fleet data back into post-training and rolls out improved policies safely.
Own program execution end to end. Set strategic direction, identify resource needs, and answer for delivery against company goals and the roadmap. Challenge assumptions at the org level and build a results-driven culture.
Communicate program status, risks, trade-offs, and decisions to executive stakeholders. Make compute, data, and model trade-offs clear to leadership, and state requirements, goals, and milestones clearly across teams.
Mentor TPMs and engineers. Define the frameworks, tooling, and operating rhythms that let research move quickly without losing rigor, and raise the technical bar of the program management function.
Requirements
Have led large model training programs, including pretraining, mid-training, fine-tuning, and distillation. Understand how runs are planned, resourced, monitored, and recovered after failure.
Understand GPU clusters in cloud and on-premises settings, including schedulers, distributed training, high-speed interconnects, and storage. Have driven capacity planning, procurement, and cluster bring-up.
Read training code, interpret loss curves and evaluation results, and diagnose data and infrastructure problems alongside the team. Be comfortable with Python and a modern framework such as PyTorch or JAX.
Have worked with robot learning data: teleoperation, demonstration capture, simulation and synthetic data, annotation, and quality control at scale.
Have built evaluation frameworks and release criteria for ML models, ideally covering both simulation and real-world testing.
Know how models move from training to on-device inference, including optimization for embedded and edge AI hardware.
Have a record of managing cloud providers, hardware vendors, and data partners, including negotiating capacity and resolving cross-team dependencies.
Plan around research uncertainty using experiments, decision points, and gates instead of fixed deliverables, while holding infrastructure and data work to firm commitments.
Have allocated scarce compute and data across competing teams, aligned research, engineering, and executive leadership, and made hard prioritization decisions.
Turn company strategy into compute, data, and model roadmaps, and shape those roadmaps with data-driven insight.
Hold a bachelor's degree in computer science, electrical or computer engineering, or a related technical field.
Have 10+ years in technical program management, ML engineering, or ML infrastructure, including leading large-scale training or compute programs at senior, staff, or principal level.
Preferred: experience at a frontier AI lab, hyperscaler, or robotics foundation model company.
Preferred: experience setting up or scaling GPU clusters or data center capacity.
Preferred: direct experience with robot learning, vision-language-action models, or large-scale robot data programs.
Preferred: a master's or PhD in computer science, machine learning, or a related field.
Physical requirements: long periods sitting at a desk using a computer, occasionally lifting 15 pounds, reading printed material and screens, and hearing and speaking to communicate.
This is a direct hire. Agency submissions are not accepted.
Conditions
Direct hire only. Agencies should not submit candidates.