Research Member of Technical Staff – Training Systems

About the role

Rhoda AI is building generalist intelligent robots and needs a senior ML Training Systems Engineer to own end‑to‑end training performance. You will define how large‑scale multimodal models train, driving efficiency, scalability, and correctness across thousands of GPUs.

Key responsibilities

• Own training performance end‑to‑end, diagnosing and improving large‑scale multimodal training (vision, video, proprioception, actions, language).
• Build systematic performance attribution, scaling curves, and bottleneck identification to drive gains in distributed, compute, and memory efficiency.
• Design and evolve parallelism strategies (data, tensor, pipeline, sharding, hybrid) and improve execution through communication scheduling and runtime optimizations.
• Partner with research scientists to translate model innovations into scalable implementations and collaborate with infrastructure/SRE teams on cluster‑level efficiency.

Required profile

• Proven track record improving large‑scale distributed training performance.
• Deep hands‑on experience with modern ML stacks, especially PyTorch (JAX a plus).
• Strong understanding of data/tensor/pipeline parallelism, sharded training (FSDP/ZeRO), and GPU cluster scaling.
• Exceptional debugging and measurement skills to turn performance issues into validated improvements.

Required skills

• PyTorch
• JAX (optional)
• CUDA
• Triton
• Distributed training frameworks and GPU clusters
• FSDP / ZeRO‑style sharding
• Data, tensor, pipeline parallelism