World Labs
Performance Engineer (Inference, Training & GPU)
$200K–$300K
San Francisco · On-site · Full-time · Posted May 1, 2026
- Visa sponsorship
- Visa sponsorship is offered; success is not guaranteed.
About the job
Optimize World Labs’ AI models for training and production inference by improving GPU kernels, utilization, serving speed, and numerical correctness.
What you'll do
- Optimize inference and serving performance, including latency, throughput, batching, caching, and scheduling.
- Tune GPU kernels and improve training throughput and GPU utilization.
- Build profiling, performance-modeling, and observability tools; partner with researchers to productionize models.
What they ask for
- Strong performance-engineering foundations, including profiling, roofline analysis, and root-cause investigation.
- Deep GPU programming and optimization experience with CUDA and/or Triton.
- Hands-on experience optimizing inference and serving for large models and optimizing training performance.
- Working knowledge of PyTorch and/or JAX; strong Python proficiency and ability to use C++/CUDA as needed.
Nice to have
- Experience at an AI lab or ML-native company, or serving generative, diffusion, video, or 3D/spatial models.
- Experience with low-precision numerics, distributed systems, multi-accelerator systems, or GPU performance observability.
Pay
$200K – $300K (base salary)
Tools
CUDA, Triton, FP8/INT8, PyTorch, JAX, torch.compile, XLA, NCCL
AI focus
Accelerate training and inference for large generative world models, including production serving.
Model-written summary · the original posting is authoritative