Deep dives into CUDA optimization, deep learning systems, and lessons learned building production AI at scale.
From 1.75 to 22.6 TFLOPS - a deep dive into CUDA tensor cores, PTX instructions, and the optimization techniques that let my implementation outperform PyTorch's cuDNN at long sequence lengths.
A comprehensive guide to building autonomous AI systems - from core architecture modules and design patterns to framework selection and production deployment strategies.
How I built a real-time multi-person pose tracker using YOLOv8-Pose, TensorRT, and a fully GPU-native ByteTrack implementation. Achieves 420 FPS on RTX 3080 Ti with zero CPU bottleneck.
I love talking about CUDA, deep learning systems, and high-performance computing. Let's connect.
Get in Touch