Classytic

Build a GPT one part at a time: a neuron, a trained network, convolution, attention, a transformer block, and the steps that turn a raw model into an assistant. A tiny name generator trains live in your browser and grows up with you.

Radio, Wi-Fi, mobile and satellite are one phenomenon at different frequencies, and the frequency decides almost everything. Work out why a remote must be pointed and Wi-Fi need not be, then stand a router in a real flat and find the rooms it cannot reach.

Assume one thing, that light travels at the same speed for everyone, and follow it until clocks disagree, simultaneity dissolves and metre rules shrink. Every result is derived on screen, and the course ends on the muons that prove it.

Start by fitting a straight line and finish by predicting the next token, using the same idea the whole way. Every model here is small enough to check by hand, and every chapter ends where a real production system actually stands today.

Build circuits from the gate up, then give them memory. Simplify with Karnaugh maps, wire an adder, watch two gates hold a bit, and design a machine that finds a pattern. Follows the first digital logic course at universities such as NSU (CSE231) and BRAC (CSE260).

Go inside the crystal to see where carriers come from, why a junction conducts in only one direction, and how a third terminal takes control of a current. Built on simulations, not slides.

Build a network from the cable up. Wire a plug, watch a switch learn, move a prefix line and see two machines stop being neighbours, then work out exactly who can read your traffic and what a VPN does about it.

Designing a service and running one are different jobs. Replace drifting host installations with immutable images, hand a fleet a desired state and let controllers reconcile it, then judge the result by what customers actually experienced rather than by what the dashboards say.

Build the portable mental model underneath AWS, Google Cloud and Hetzner: publish one controlled service endpoint, diagnose every boundary, distribute traffic only to healthy backends and scale before overload.

Every method here reports its own price while it runs. Count the comparisons, watch the counter climb, and let Big-O arrive as a summary of numbers you have already seen rather than a table to memorise.

A generative model in production is a capacity problem before it is anything else. Budget the memory a model and its KV cache actually occupy, batch requests without starving any of them, and scale against demand rather than pretending accelerators are infinite.