Writing

Essays on systems, ideas, and messy reality, from someone still figuring things out.

4 posts
2026
Fastest Matrix NMS in the West (Part 2): Going Down to CUDA
ArchivedAICS

A faster Matrix NMS implementation using custom CUDA kernels.

Sep 7, 2026
2023
Fastest Matrix NMS in the West (Part 1): A Naive Implementation
ArchivedAI

Introduction to Matrix NMS, plus a naive implementation and its problems.

Feb 11, 2023
2022
From ResNet to ConvNeXt (Part 2): Modernizing a Vanilla ResNet
ArchivedAI

Progression from ResNet to ConvNeXt, step by step explained with code in PyTorch.

Apr 4, 2022
From ResNet to ConvNeXt (Part 1): ResNet with Better Training Techniques
ArchivedAI

Motivation of the ConvNeXt paper and a refresher on ResNet in PyTorch.

Mar 22, 2022