Friend Links

Thanks for stopping by đź‘‹
Here you’ll find a curated collection of my Medium articles — shared through friend links so you can read them without a paywall.

If you enjoy the content and want to support my work, a few small actions go a long way:

  • đź‘Ť Leave a like or highlight sections you find useful
  • đź’¬ Share your thoughts or questions in the comments
  • đź”– Bookmark articles you want to revisit
  • 👤 Follow me on Medium to stay updated
  • đź“– Spend a moment reading — thoughtful engagement helps the work reach more people
  • đź”— Share articles with friends or colleagues who might find them helpful

Your support helps me continue creating deep, practical content on machine learning, computer vision, and efficient engineering.

Articles

  • The Tricks That Make Production 3DGS Fast (Even If Ours Isn’t)
  • Splat Your Own Gaussians - From Circles to Ellipses
  • Circles Are Not Gaussians (But Let’s Pretend They Are)
  • I Built the Slowest 3D Gaussian Splatting Renderer… On Purpose
  • Still Avoiding einsum()? It’s Time to Fix That
  • Mastering NumPy - Manual Metadata Manipulation for Memory-Efficient Arrays
  • The Power of Views - How NumPy Avoids Copies and Saves Memory
  • Why NumPy Arrays Are So Fast (And How They Really Work)
  • FlashAttention — Visually and Exhaustively Explained
  • FlashAttention from First Principles
  • Vision Mamba - Like a Vision Transformer but Better
  • Here Comes Mamba - The Selective State Space Model
  • Structured State Space Models Visually Explained
  • Towards Mamba State Space Models for Images, Videos and Time Series
  • The Rise of Diffusion Models - A new Era of Generative Deep Learning
  • Depth Anything - A Foundation Model for Monocular Depth Estimation
  • Turn Yourself into a 3D Gaussian Splat
  • DINO - A Foundation Model for Computer Vision
  • Segment Anything - Promptable Segmentation of Arbitrary Objects
  • BYOL -The Alternative to Contrastive Self-Supervised Learning
  • GLIP - Introducing Language-Image Pre-Training to Object Detection
  • The CLIP Foundation Model
  • Implement Multi-GPU Training on a single GPU
  • Fourier CNNs with Kernel Sizes of 1024x1024 and Larger
  • MLP Mixer in a Nutshell
  • Create your own GPU accelerated Jupyter Notebook Server with Google Colab using Docker
  • Accelerated Distributed Training with TensorFlow on Google's TPU
  • Speed up your Training with Mixed Precision on GPUs and TPUs in TensorFlow