Performance Engineer: GPU Kernel & Inference Optimize

WORLD LABS; San Francisco (CA); On-site; USD 200,000 - 300,000; Full time

Job summary

World Labs in San Francisco is hiring a Performance Engineer to accelerate training and serving of large world models. You will identify bottlenecks across kernels, serving paths, and GPUs, then implement concrete improvements that raise throughput and lower latency.

You’ll work end-to-end, from CUDA kernels to fleet-wide serving, partner with researchers, and own numerical correctness across precision and hardware changes.

Qualifications

• Strong fundamentals in profiling, roofline analysis, latency/throughput optimization, and root-cause investigation.
• Deep GPU programming and optimization experience (CUDA and/or Triton) with kernel-level tuning.
• Experience optimizing inference/serving for large models and mixed/low precision execution.
• Knowledge of ML framework internals (PyTorch and/or JAX).

Responsibilities

• Optimize inference and serving end to end for production-scale models.
• Write and tune GPU kernels (CUDA, Triton) and drive memory/bandwidth optimization.
• Improve training throughput and GPU utilization with parallelism and mixed precision.
• Build performance models, profiling workflows, and observability across the stack.
• Ensure numerical correctness across precision and hardware changes.

Skills

Profiling & analysis

GPU optimization

CUDA

Triton

Python

C++/CUDA

Distributed systems

PyTorch/JAX

Tools

CUDA

Triton

PyTorch

JAX

Job description

World Labs in San Francisco is hiring a Performance Engineer to accelerate training and serving of large world models. You will identify bottlenecks across kernels, serving paths, and GPUs, then implement concrete improvements that raise throughput and lower latency.

You’ll work end-to-end, from CUDA kernels to fleet-wide serving, partner with researchers, and own numerical correctness across precision and hardware changes.