Paul D.

Hi there ! I'm Paul, a ML Performance Engineer. I specialize in both deploying and optimizing AI systems for efficient inference and training on modern GPU. • Skills : - Core Languages : CUDA C++, Python, Assembly, Golang - High-Perf ML : PyTorch, JAX, vLLM, Nsight Compute, Triton, CUTLASS, MLIR, ONNX - MLOps/Infra : Linux, SLURM, Docker, Git, DeepSpeed, FSDP, pytest • Recent hands-on work : - Custom CUDA fused kernels with BF16 tensor cores, FP32 accumulation: 7× speedup over PyTorch baseline - Production multi-model vLLM stack on a 12× RTX 4090 / dual EPYC cluster I configured from bare metal - ZeRO-2/3 benchmarking on Qwen2.5 7B with MFU tracking and Nsight Compute profiling