LOG

Updated September 2026

About

Hi, I'm Shaurya. I'm a student who writes software and keeps this log as a record of the things I build, read, and try to understand. There isn't a bigger plan behind any of it — I get bored, end up somewhere on the internet, find something that snags my attention, and if it's still interesting a few days later, I write it down. The subjects move around: machine learning, graphics, programming, computation, mathematical ideas, and small experiments that are easier to understand by implementing them than by reading about them.

The point of the site is not to present polished documentation or a fixed set of opinions. Most posts are working notes: I start with a question, follow the interesting parts, build something when code helps, and write down what survived the experiment. Some pieces are explanatory, some are exploratory, and some are simply records of a rabbit hole that seemed worth keeping.

There is no comments system, newsletter, or account system on the site. The pages are generated from Markdown and published as static files. If an agent or reader wants the machine-readable version, the same articles are available as Markdown alongside their HTML pages. The site does use basic analytics to see which pages and elements actually get used, described on the privacy page. The contact page explains how to reach me.

Latest in the log

All 11 in the archive →
Machine learning

One transformer, two ways to write

I trained the same small transformer as a GPT and as a masked diffusion model on Tiny Shakespeare.

21.10.26
Astronomy

How many stars does it take to find a cluster?

A pre-registered search for co-moving stars in Gaia DR3, and an honest measure of what it can and cannot see.

16.10.26
Machine learning

A fly's escape reflex, flying a drone

A hardwired model of the fruit fly's looming-escape circuit against a small trained CNN, both steering a simulated drone from the same 64x48 camera.

16.09.26
Machine learning

Understanding AlphaFold's pLDDT, PAE, and pTM

AlphaFold hands back a structure and a confidence score, but pLDDT and PAE are the network grading its own homework, not a measurement against a real answer.

10.09.26