GPU Optimization Engineer at High-growth generative AI startup

You will lead the optimization of real-time speech model architectures, ensuring maximum throughput and minimal latency across diverse GPU hardware. This role involves deep architectural tuning, custom kernel development, and hardware-specific profiling to squeeze every millisecond of performance out of state-of-the-art models. You will directly impact the reliability and speed of cutting-edge AI inference stacks.

High-growth generative AI startup

• Design and implement custom CUDA, Triton, or Tinygrad kernels for performance-critical model sections.
• Profile end-to-end inference workloads using tools like Nsight to identify and resolve memory bandwidth and kernel bottlenecks.
• Partner with research and infrastructure teams to perform operator fusion, graph optimization, and kernel-level scheduling.

• Possesses a Master’s or PhD in GPU Programming with 3-5 years of specialized experience in hardware-level optimization.
• Demonstrates deep mastery of GPU architecture, including SMs, memory hierarchy, occupancy tuning, and kernel debugging.
• Has extensive hands-on experience with PyTorch, TensorRT, and various model architectures like transformers and diffusion blocks.

===

• Influence the performance of state-of-the-art real-time audio models at a deep architectural level.
• Join a well-funded team backed by top-tier VCs working on the frontier of generative speech technology.
• Work across diverse hardware backends, porting models from NVIDIA to AMD and emerging edge accelerators.