Why this post exists

This is a technical writeup stored as a single MDX file under content/posts/. You can use normal markdown: headings, lists, and images. (Revision B: tightened intro after review.)

A little math

When we minimize a scalar loss $L(\theta)$, a vanilla update is:

Inline math works too: the learning rate $\eta$ controls step size.

Interactive diagram

Drag the slider and press Gradient step to watch $\theta$ move along a simple 1D loss curve.

Writing process

Commit this file as you draft. The history control on this page reads git log for this path so you can scrub the timeline and see earlier wording.