CUDA Kernel Engineer at Pragmatike

Department: Work with our Clients

Setup: On-site
Location: Cambridge, England
Level: senior
Posted: Apr 22, 2026

Description

Pragmatike is hiring a CUDA Kernel Engineer to design, implement, and optimize custom NVIDIA CUDA kernels for high-throughput AI systems serving Fortune 500 customers. The role focuses on GPU occupancy, memory throughput, warp efficiency, profiling, memory pipelines, PCIe transfers, and kernel libraries, with collaboration across AI systems, model acceleration, and distributed systems teams. Candidates need hands-on experience building CUDA kernels from scratch, strong knowledge of GPU architecture and CUDA runtime APIs, and familiarity with C++ and GPU debugging tools.

Required Skills

CUDA kernels
GPU architecture
Memory coalescing
Warp divergence
C++
CUDA APIs
Nsight Compute
NVLink

Additional Information

Required Previous Experiences: Hands-on experience developing and optimizing NVIDIA CUDA kernels from scratch
Proven track record building NVIDIA CUDA kernels from scratch
Preferred Qualifications: Experience with multi-GPU or distributed GPU systems
Background in GPU acceleration for machine learning frameworks or high-performance computing workloads
Knowledge of model inference optimization
Exposure to compiler-level optimization or PTX/SASS analysis
Startup experience or comfort working in fast-moving, ambiguous environments