System Design

179 boards, 46 lessons, 631 review questions and 103 narrated videos, in 24 chapters. Read them in order. Each chapter builds on the one before.

Your progress0 of 46 read

Start here → The chai stall that got famous (load balancing)

WatchAmazon9 Discord9 Figma9 Google Search9 Netflix9 Slack9 Stripe9 Uber9 WhatsApp9 YouTube9 All 103 videos →

Contents

Part one · Foundations

1The stall gets famous5 lessons
One chai stall, and everything that breaks as it gets famous: more stalls, a slate by the stove, a rail of order slips, copy clerks, and a ledger cut in two.
▶ Watch the video
The Chai Stall That Got Famous The Slate Next to the Stove The Order Slip Rail The Copy Clerks Splitting the Ledger
2When the stalls must agree5 lessons
Two books that should say the same thing and a wire that gets cut: CAP, who holds the pen, one transfer across two ledgers, the double charge, and the log everyone reads.
after The stall gets famous▶ Watch the video
The Cut Phone Line Who Holds the Pen The Two Ledgers Must Agree Charged Twice The Ledger Everyone Reads
3Holding up under load4 lessons
What one wala does while the chai brews, then rate limits, turning people away on purpose, and the arithmetic of the nines when thirty servers answer one request.
after The stall gets famous▶ Watch the video
The Bouncer at the Door
4The parts, properly9 lessons
The pieces again at real depth: the map of all five, then choosing a database and guarding the last item, caching and the branch near your house, counting a crowd, the contract an API makes, keeping a line open, and a lock on every door.
after The stall gets famous▶ Watch the video
Data Kya Hai (Why It Matters)The Right Book for the Job The Slate, Properly The Contract at the Counter

Part two · Build

5The interview, assembled2 lessons
The six steps every design question runs through, then four whole systems designed end to end in one night.
after Holding up under load, The parts, properly
- 01Design the Whole Thing (the interview, assembled)11 Q
- 02Four Systems in One Night (worked designs)9 Q
Design the Whole Thing Four Systems in One Night
6The stall, built3 lessons
The spine as running Go: a gate, a slate and a rail, then copies and shards with a stale read on purpose, then a load generator that finds the knee.
after The interview, assembled

Part three · Depth

7Amazon1 lesson
From a memo that became a one-click cart, through blast radius and eleven-nines reliability, to the typo that took down half the internet.
after The stall gets famous▶ Watch the video Amazon case study →
- 01Amazon - the case study22 Q
The Map The Memo That Became a Business The Cart That Cannot Say No Blast Radius The Coffee Urn Eleven Nines Chaotic Storage The Typo That Broke the Internet Everything
8Claude1 lesson
The request path through an orchestrator and context window, RAG and the agent loop, the inference fleet, and the training and eval pipeline behind it.
after The stall gets famousClaude case study →
- 01Claude - the case study25 Q
The Map The Request Path Orchestrator & Context Window RAG & The Agent Loop Inference Fleet Internals Safety & Guardrails Training & Eval Pipeline Reliability, Cost & Multi-Region Everything
9Cloudflare1 lesson
One IP serving 330 cities, isolates instead of containers, an unpatchable Spectre, and 434 years of TLS handshakes saved by session resumption.
after The stall gets famousCloudflare case study →
- 01Cloudflare - the case study24 Q
The Map One IP, 330 Cities The Isolate You Cannot Patch Spectre One Object, One Place, One Thread Global In Seconds The Attack Divides Itself 434 Years of Handshakes Everything
10Discord1 lesson
The gateway that holds millions of sockets open, the message-store rewrite, voice and video, and fan-out at the limit of a single channel.
after The stall gets famous▶ Watch the video Discord case study →
- 01Discord - the case study20 Q
The Map The Gateway The Message Store Saga Data Services & Super-Disk Voice & Video Fanout at the Limit Search The API Plane Everything
11Dropbox1 lesson
Why sync is not copying - three trees and a merge base, the rewrite that worked, a fuzzer that found what code review didn't, and disks that lie.
after The stall gets famousDropbox case study →
- 01Dropbox - the case study28 Q
The Map Sync Is Not Copying The Rewrite That Worked Three Trees and a Merge Base The Fuzzer Found It Leaving the Cloud The Disks Are Lying To You The Blocks Don't Know Your Name Everything
12Figma1 lesson
A browser inside a browser, a way of merging two people's edits that is neither of the two textbook answers (OT, CRDT), one process per document, and undo when someone else already moved it.
after The stall gets famous▶ Watch the video Figma case study →
- 01Figma - the case study33 Q
The Map A Browser Inside a Browser Neither OT nor CRDT One Process Per Document Drawing a Letter Without a Font Engine Nine Months to Shard One Table Undo, When Someone Else Moved It The Plugin That Can See Everything Everything
13Google Maps1 lesson
The road graph and routing algorithm underneath the blue line, traffic-aware alternates, tile serving, and the geospatial index that finds nearby places.
after The stall gets famousGoogle Maps case study →
- 01Google Maps - the case study21 Q
The Map Road Graph & Routing Traffic-Aware & Alt Routes Serving the Map (Tiles)Geospatial Index & Places Real-Time Traffic ETA & Prediction (ML)Map Data Pipeline & Scale Everything
14Google Search1 lesson
The crawler, the inverted index it feeds, query serving and ranking, and how freshness and tiered serving keep an index of the whole web current.
after The stall gets famous▶ Watch the video Google Search case study →
- 01Google Search - the case study17 Q
The Map The Crawler The Inverted Index Query Serving Ranking Freshness & Tiered Serving Storage & Compute Infra Beyond Ten Blue Links Everything
15Instagram1 lesson
How three engineers shipped it, then sharding and IDs, the celebrity problem in a feed, and Python holding at three billion requests a day.
after The stall gets famousInstagram case study →
- 01Instagram - the case study19 Q
The Map The 3-Engineer Stack Sharding & IDs Feed & the Celebrity Problem The Data Layer Saga Python at 3 Billion Stories & Explore ML Two Continents, One App Everything
16Netflix1 lesson
A read-heavy streaming giant, end to end - encode, cache, recommend.
after The stall gets famous▶ Watch the video Netflix case study →
- 01Netflix - the case study18 Q
The Map Open Connect CDN Adaptive Bitrate Streaming The Encoding Pipeline Playback Control Plane Recommendations & Personalization Microservices & Resilience Data Platform & A-B Everything
17OpenRouter1 lesson
One API in front of every model - routing, wiretapping traffic it has to bill for, counting tokens it doesn't own, and the one thing a reseller can't copy.
after The stall gets famousOpenRouter case study →
- 01OpenRouter - the case study16 Q
The Map One API Every Model The Router The Wiretap That Must Bill Three Nets Under One Tightrope Counting Tokens You Dont Own A Toll Booth on a Road They Dont Own The One Thing You Cant Copy Everything
18Slack1 lesson
The monolith that never died, the thundering herd on reconnect, a shard key that lied, and a queue that could not drain itself.
after The stall gets famous▶ Watch the video Slack case study →
- 01Slack - the case study26 Q
The Map The Monolith That Never Died The Thundering Herd The Message Path The Shard Key That Lied The Queue That Could Not Drain One Channel, Two Owners Search, Ranking and the Model That Must Not Learn Everything
19Stripe1 lesson
Idempotency as the API's first principle, a state machine for moving money, a double-entry ledger, and the card-network rails underneath all of it.
after The stall gets famous▶ Watch the video Stripe case study →
- 01Stripe - the case study23 Q
The Map The API & Idempotency Money Movement State Machine The Double-Entry Ledger Card Network Rails Reliability & Exactly-Once Data Infra & Scale Beyond Payments Everything
20Twitter (X)1 lesson
Timeline fan-out on write versus read, posting and reading at the same instant for millions, and the social graph that decides who sees what.
after The stall gets famousTwitter (X) case study →
- 01Twitter (X) - the case study18 Q
The Map Timeline Fan-out Posting a Tweet Reading the Home Timeline The Social Graph Search & Trends Storage & Caching Scale, Reliability & Delivery Everything
21Uber1 lesson
A location firehose feeding a geo-index, dispatch and matching, and surge pricing as a live market clearing supply against demand.
after The stall gets famous▶ Watch the video Uber case study →
- 01Uber - the case study18 Q
The Map Location Firehose & Geo-Index Dispatch & Matching Surge Pricing The Trip State Machine Payments & Money Maps, Routing & ETA Scale, Cells & Data Platform Everything
22WhatsApp1 lesson
The persistent connection every phone holds open, a message's trip from Mumbai to Virginia, and end-to-end encryption that keeps the server from reading it.
after The stall gets famous▶ Watch the video WhatsApp case study →
- 01WhatsApp - the case study20 Q
The Map The Persistent Connection The Message Journey (India to US)Real-Time & Durability The Database Layer Groups & Fan-out End-to-End Encryption Scale, Presence & The Erlang Story Everything
23YouTube1 lesson
A nine-engineer beginning, Vitess sharding MySQL underneath it, the transcode factory behind a billion hours watched a day, and Content ID fingerprinting at that scale.
after The stall gets famous▶ Watch the video YouTube case study →
- 01YouTube - the case study17 Q
The Map The 9-Engineer Miracle Vitess The Transcode Factory A Billion Hours a Day Recommendations Content ID The Cheat & Procella Everything
24Zomato1 lesson
Search and discovery, the order lifecycle, assignment and batching of delivery partners, live tracking, and a three-sided payment system underneath it.
after The stall gets famousZomato case study →
- 01Zomato - the case study18 Q
The Map Search & Discovery The Order Lifecycle Assignment & Batching Live Order Tracking 3-Sided Payments The Restaurant Platform Scale, ETA & ML Everything

Borrowed from other tracks (13)

You don't need to take those tracks. These are the only lessons this track uses from them, and each is linked where it is needed.