Pragmatike is hiring a CUDA Kernel Engineer on behalf of a fast-growing AI startup founded by MIT CSAIL researchers. You will design, implement, and optimize custom CUDA kernels from scratch for NVIDIA GPUs, directly powering the high-throughput AI systems used by Fortune 500 clients.

What You'll Do

Design, implement, and optimize custom CUDA kernels for NVIDIA GPUs, focusing on maximizing occupancy, memory throughput, and warp efficiency.
Profile GPU workloads using tools such as Nsight Compute, Nsight Systems, nvprof, and CUDA‐MEMCHECK.
Analyze and eliminate performance bottlenecks including warp divergence, uncoalesced memory access, register pressure, and PCIe transfer overhead.
Improve GPU memory pipelines and ensure proper memory coalescing.
Collaborate closely with AI systems, model acceleration, and backend distributed systems teams.
Contribute to GPU architecture decisions, kernel libraries, and internal performance-engineering best practices.

What We're Looking For

Proven track record building NVIDIA CUDA kernels from scratch, not just calling existing libraries.
Strong ability to optimize kernels using tiling strategies, occupancy tuning, shared memory design, and warp scheduling.
Deep understanding of CUDA threads, warps, blocks, and grids, GPU memory hierarchy and memory coalescing, as well as warp divergence.
Experience diagnosing PCIe bottlenecks and optimizing host-device transfers using pinned memory, streams, and batching.
Familiarity with C++, CUDA runtime APIs, and GPU debugging and profiling tooling.

Nice to Have

Experience with multi-GPU or distributed GPU systems such as NCCL or NVLink.
Background in GPU acceleration for ML frameworks or HPC workloads.
Knowledge of model inference optimization with TensorRT, CUDA Graphs, or CUTLASS.
Exposure to compiler-level optimization or PTX/SASS analysis.
Startup experience or comfort working in fast-moving, ambiguous environments.