Under the Hood

About

What actually happens under the hood.

Most learning material for backend and system design lives at two unhelpful extremes: shallow tutorials that tell you which command to type, and dense textbooks and papers that assume you already have the map. Under the Hood is an attempt at the missing middle — explanations that start from what a working developer already knows and dig down to the real mechanics, one honest layer at a time.

Two halves that check each other

The curriculum goes from first principles: what really happens when you call fetch(), all the way down through DNS, TCP, TLS, and HTTP; how a database actually executes a query — pages, B-trees, the write-ahead log, MVCC; and then the building blocks, consistency trade-offs, APIs, distributed patterns, and worked interview designs that build back up. Every lesson ends with self-check questions and a curated set of deeper references, each with a reason to read it.

The other half, Fable, Deconstructed, keeps the theory honest. Fable is a real, shipped group-expenses app, and each chapter takes a piece of it — the money-exact data model, the settlement state machine, real-time chat, authentication, the infrastructure migration that started slow and got fixed — and explains the actual decision, the trade-off, and, where it happened, the production bug. Every curriculum topic links to where it lives in Fable, so no concept stays abstract. The theory says how things work; Fable shows what it costs to get them right.

The through-line

If the whole site argues one thing, it's this: the hard engineering skill is not assembling sophisticated systems, but knowing exactly which sophistication to refuse — and leaving the seam where you'd add it later. Fable runs on a single server on purpose, and every chapter is a worked example of choosing the simplest design that meets the requirement and naming the trade-off out loud.

Who & how

Written by Krishna Agarwal, a frontend developer who wanted to genuinely understand the systems underneath. It's built as a fully static site — Fumadocs on Next.js 15, hand-rolled CSS, no analytics, no accounts, no tracking — and the content is fact-checked against Fable's real source code rather than written from memory.

Start with the curriculum →