Deep in the Stack

Unfinished series

Exploring AI infrastructure, one layer at a time

How This Started

I was bored. Just scrolling X, watching the same surface-level AI news over and over, new model drops, benchmark scores, the usual hype cycle. Struck my mind that I'm just consuming some info superficially, idk what is happening under the hood.

Was having a conversation with a junior and a friend. Gave a bunch of resources, told me to dive into Andrej Karpathy's playlist on YouTube.

So I did. Randomly, almost on a whim. I started watching, started coding, started making a plan. v0 plan is:

• Build micrograd + understand backprop
• Build a transformer from scratch
• Write a GPT tokenizer
• Build a mini GPT
• Read around vLLM
• Eventually, build a mini vLLM

This series is the documentation of that journey. Every step, every build, every misunderstanding along the way.