ML Performance Engineer: Distributed Training & Inference

Veeda AI; Seattle (WA); On-site; USD 180,000 - 260,000; Full time

Job summary

Veeda AI is seeking a highly skilled engineer to advance distributed training and high-performance inference for multimodal world models. You will own multi-node throughput, profile GPU utilization, and optimize kernels and CUDA/Triton code across NVLink fabrics.

Candidates should have deep PyTorch expertise, Python and C++/CUDA fluency, and experience with frameworks like FSDP2, Megatron-Core, DeepSpeed. You will work in a fast-moving research environment to push physical AI capabilities

Qualifications

• Bachelor's degree in Computer Science, Computer Engineering, or related field.
• Experience with PyTorch and a large-scale parallelism stack (FSDP2, Megatron-Core, TorchTitan, or DeepSpeed).
• Fluency in Python and C++/CUDA with the ability to predict kernel stalls before profiling.
• Experience profiling live training/inference with Nsight Systems or PyTorch profiler.
• Expertise in low-precision numerics, kernel authoring, or fault diagnosis, with credibility in others.
• Experience optimizing training and inference on very large GPU clusters, including topology-aware placement and scaling efficiency.

Responsibilities

• Own step time and model FLOPs utilization for multi-node video world model training, selecting parallelism strategies.
• Profile end-to-end latency, GPU utilization, memory pressure, and kernel efficiency; improve throughput with advanced techniques.
• Address precision and numerical stability challenges across FP16/FP8 and related formats in video tokenizer and VAE layers.
• Write and tune CUDA and Triton kernels to optimize attention and kernel performance beyond standard PyTorch implementations.
• Tune NCCL collectives and overlap across NVLink domains to maximize throughput.
• Develop fault diagnostics and asynchronous checkpointing to minimize interruption costs.

Skills

Python

C++

CUDA

PyTorch

Distributed training

Performance profiling

Education

Bachelor's degree in Computer Science or related field

Tools

PyTorch FSDP2

Megatron-Core

TorchTitan

DeepSpeed

Nsight Systems

NCCL

Job description

Veeda AI is seeking a highly skilled engineer to advance distributed training and high-performance inference for multimodal world models. You will own multi-node throughput, profile GPU utilization, and optimize kernels and CUDA/Triton code across NVLink fabrics.

Candidates should have deep PyTorch expertise, Python and C++/CUDA fluency, and experience with frameworks like FSDP2, Megatron-Core, DeepSpeed. You will work in a fast-moving research environment to push physical AI capabilities