ML Systems Engineer — Performance & Scale

Veeda Innovation; Northern (KY); Hybrid; USD 150,000 - 230,000; Full time

Job summary

Veeda AI is seeking a Member of Technical Staff - ML Performance to optimize distributed training and inference for large-scale video models. You will drive throughput, profile bottlenecks, and implement kernel-level enhancements across PyTorch, CUDA, and specialized parallelism stacks.

The role requires deep experience with FSDP2, Megatron-Core, TorchTitan, or DeepSpeed, plus strong Python/C++ skills. Location in the US is expected, with a focus on scalable acceleration across GPU clusters.

Qualifications

• Experience with PyTorch and large-scale multi-node parallelism stacks on real workloads.
• Fluency in Python and C++/CUDA, with kernel performance intuition.
• Proven ability to profile training/inference and translate traces into gains.

Responsibilities

• Own training throughput and mix parallelism for multi-node video world model training.
• Profile latency, GPU utilization, memory pressure, and improve throughput with optimizations.
• Tune kernels and integrations to reduce step time and enable efficient attention.
• Optimize communication and overlap across NVLink and NCCL for scalability.
• Develop fault diagnostics and asynchronous checkpointing to reduce downtime.

Skills

PyTorch

Python

C++/CUDA

FSDP2

Megatron-Core

DeepSpeed

TorchTitan

Profiling

Education

Bachelor's degree in Computer Science or related field

Tools

Nsight Systems

PyTorch Profiler

CUDA

NCCL

torch.compile

Triton

CuTe-DSL

Job description

Veeda AI is seeking a Member of Technical Staff - ML Performance to optimize distributed training and inference for large-scale video models. You will drive throughput, profile bottlenecks, and implement kernel-level enhancements across PyTorch, CUDA, and specialized parallelism stacks.

The role requires deep experience with FSDP2, Megatron-Core, TorchTitan, or DeepSpeed, plus strong Python/C++ skills. Location in the US is expected, with a focus on scalable acceleration across GPU clusters.