Blog

01 // ENTRIES

1. 01

   xSLAM: Building a Real-Time SLAM Pipeline in the Browser_

   2025-05-12

   Why I built a fully client-side 3D Gaussian Splatting and Graph SLAM visualizer, and how WebAssembly enables sub-10ms sensor-fusion loops directly in the browser with zero backend dependencies.

2. 02

   copypaste.fyi: Secure, Deterministic Clipboard Relay_

   2024-11-04

   The motivation behind copypaste.fyi: designing a low-latency edge-native relay for developers who live in tmux, and engineering short-lived encrypted token exchanges.

3. 03

   Writing Custom PyTorch C++ Extensions for Real-Time Inference_

   2018-03-22

   How to build, compile, and deploy custom CUDA operators as PyTorch C++ extensions — bridging the gap between Python prototyping and sub-millisecond production inference on embedded GPUs.

4. 04

   Lessons from Training GANs: Stability, Mode Collapse, and DCGAN_

   2016-08-07

   Hard-won lessons from months of training generative adversarial networks — understanding the minimax game, diagnosing mode collapse, and the architectural tricks that finally made image synthesis work.

5. 05

   Implementing ConvNets from Scratch in C++_

   2013-09-18

   A retrospective on building a minimal convolutional neural network in raw C++ before frameworks existed — forward passes, hand-rolled backpropagation, and the GPU compute revolution that changed computer vision forever.