Research Scientist / Engineer – Performance Optimization

Highlights

• Make Luma's multimodal models fast for training and large-scale deployment.
• Develop fused kernels that maximize hardware utilization without quality loss.
• Build monitoring and automation that keeps performance gains from regressing.

Ideal candidate

Deep performance specialist who lives in profilers and loves squeezing maximum utilization from hardware. You care about efficient large-scale training and deployment of multimodal transformers without sacrificing quality.

Requirements

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