Research Scientist / Engineer – Performance Optimization

Highlights

• Make Luma's multimodal models fast to train efficiently and deploy at scale.
• Own end-to-end speed from diagnosis to multi-node deployment at scale.
• Build monitoring and automation that locks in performance gains.

Ideal candidate

A performance-obsessed engineer who loves squeezing every ounce of speed from hardware without sacrificing quality. You think deeply about model internals and live for finding and fixing bottlenecks at scale.

Requirements

• Expert-level Triton/CUDA programming and GPU optimization
• Strong PyTorch skills, including kernel development and custom operations
• Proficiency with profiling tools (NVIDIA Nsight, torch profiler, custom tooling)
• Deep understanding of transformer architectures and attention mechanisms