ML Systems Engineer: GPU Training & Infra

Dyna Robotics; Redwood City (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Dyna Robotics is seeking a Research Engineer to architect and build end-to-end training infrastructure for a multi-cloud GPU fleet. You will ensure GPUs are utilized efficiently, runs are reproducible, and researchers can iterate with a single command.

You will own training infrastructure end-to-end, working on scalable systems for large multimodal models and real-time robot data processing, with emphasis on performance, reliability, and deployable inference.

Qualifications

• 7+ years of engineering experience in HPC or ML infrastructure.
• Deep experience with PyTorch and distributed training frameworks (DeepSpeed, Accelerate).
• Hands-on cloud GPU management (GCP/AWS) and container orchestration (Kubernetes).
• Strong understanding of distributed systems, NCCL inter-node communication, memory management.

Responsibilities

• Scale distributed training infrastructure for large-scale GPU clusters, including sharding, checkpointing, and memory optimization.
• Build a research codebase and job scheduling system to enable fast iteration and reliable retries.
• Create high-throughput data pipelines for terabytes of multimodal robot data, ensuring GPUs are never starved.
• Develop low-latency production inference pipelines with quantization and model compilation (TensorRT, Triton).
• Profile GPU utilization and memory to optimize performance across the compute fleet.

Job description

Dyna Robotics is seeking a Research Engineer to architect and build end-to-end training infrastructure for a multi-cloud GPU fleet. You will ensure GPUs are utilized efficiently, runs are reproducible, and researchers can iterate with a single command.

You will own training infrastructure end-to-end, working on scalable systems for large multimodal models and real-time robot data processing, with emphasis on performance, reliability, and deployable inference.