Rhoda AI
Research Member of Technical Staff- Training Systems
$200K–$300K
Mountain View · Staff · Full-time · Posted Sep 18, 2026
About the job
Own and improve the performance, scalability, and efficiency of large-scale multimodal model training for robotics.
What you'll do
- Diagnose training bottlenecks and improve distributed, compute, and memory efficiency across large GPU clusters.
- Design and extend training systems, parallelism strategies, and runtime optimizations.
- Establish performance metrics, benchmarking, and regression detection, collaborating closely with researchers and infrastructure teams.
What they ask for
- Proven track record improving large-scale distributed training performance.
- Hands-on experience with modern ML stacks; PyTorch required.
- Strong understanding of data, tensor, and pipeline parallelism, sharded training (FSDP / ZeRO-style), communication overlap, and scaling across large GPU clusters.
- Strong systems reasoning, debugging, and measurement skills.
Nice to have
- JAX.
- GPU kernel or compiler-level experience, multimodal or video training, training frameworks or distributed runtimes, or cluster topology and networking familiarity.
Pay
USD 200K – USD 300K base salary
Tools
PyTorch, JAX, FSDP, ZeRO, CUDA, Triton
AI focus
Improve training systems for multimodal robotics world models involving vision, video, proprioception, actions, and language.
Model-written summary · the original posting is authoritative