GPU Performance Engineer – Member of Technical Staff

About the role

Liquid AI is building general‑purpose AI systems that run efficiently on a range of hardware, from data‑center accelerators to on‑device chips. As a Member of Technical Staff – GPU Performance Engineer, you will bridge cutting‑edge research and production by creating high‑performance CUDA kernels that accelerate training, post‑training, and inference pipelines.

Key responsibilities

• Design, implement, and ship custom CUDA kernels for novel model architectures.
• Integrate kernels into PyTorch pipelines via custom ops, extensions, and dispatch mechanisms.
• Profile and optimize end‑to‑end training and inference workflows using Nsight Systems and Nsight Compute.
• Develop correctness tests, numerics checks, and performance benchmarks to guard against regressions.
• Collaborate closely with researchers to translate promising ideas into measurable speed‑ups.

Required profile

• Hands‑on experience authoring custom CUDA kernels (beyond cuDNN/cuBLAS).
• Deep understanding of GPU architecture, memory hierarchy, occupancy, and bandwidth vs compute limits.
• Proficiency with low‑level profiling tools (Nsight Systems/Compute) and a systematic performance‑optimization methodology.
• Ability to work independently, turn ambiguous bottlenecks into concrete kernel improvements, and maintain ownership of shipped code.

Required skills

• CUDA programming
• C/C++ development
• GPU architecture knowledge
• Nsight Systems and Nsight Compute
• PyTorch custom operator integration
• Benchmarking and performance regression testing
• (Nice‑to‑have) CUTLASS, Triton kernel development