Hey, I’m Pranav. I like mathematics, programming, and writing GPU kernels. I hope you enjoy your stay.

I want to write things that at least a few people may find helpful upon reading. To that end, I will slowly polish and put out some of the things I’ve accumulated over the past 2-3 years.

For example, that time I went full extremal graph theory and proved a majority of the results myself as a challenge: not reading any proofs in texts, spending around 2-3 months, around May 2024. In fact this has been a theme in my math learning, I had no ambition or orientation towards outcomes, other than spending days and days on the same problem, that you could understand in an hour if you just read the text. I don’t regret that time, I enjoyed it quite a lot and learnt how to think, however I do wish that I was more ambitious and public with my work. I’m trying to do due diligence now, to the effort I’ve put in so far.

Little intro to groups (more coming soon).

Some linear algebra and Topology

Though I have known about basic GPU programming since August 2025, I started writing high performance GPU kernels around Jan 2026 and have found the process deeply enjoyable :)

The time between August and Jan was me lost in theory land trying to build something, will write about that soon.

my attempt at extracting signal from my very noisy brain dump of a matmul blog + kernels. I co-wrote it with Claude as my editor.

(I have severe self-diagnosed ADHD so all my thoughts are extremely compressed and hazy and I use noise and filler as a stim while decompressing).

A reasonably fast flash attention, all written in CUDA for consumer Blackwell (5090, 6000 Pro). I am doing some cleanup and final checks before I put more out. I also read the Colfax category theory of layouts paper and wrote some Python to go along with it.

On the math side, posts on topology, metric spaces, intro to homotopy theory, and category theory are in the pipeline (which is clogged btw).

One last remark, my math notes are quite dense and compact, and when I re-read it, I do put a reasonable amount of effort to reconstruct the arguments and the intuition behind it, often with some external RAM (pen-paper/blackboard) aiding me.