// System design practice

Get better at
system design.

Answer one question a day, read why the answer is right, and come back tomorrow.

Primer · How it works
How Primer works, in four steps A request goes from a user to a load balancer, to two servers, and to a database. One server stops and the request goes to the other server. Then a question card, a why panel, a streak counter, and a small lab canvas. User LB Load balancer Server A Server B Database TODAY · LOAD BALANCING Server A stops. What keeps the site up? A The load balancer sends the user to B B A bigger database C A faster phone for the user D Turn the site off and on WHY Server A stops. The load balancer sees it. It sends the user to B. The user sees no error. 2 MIN READ STREAK 1 days in a row LAB · FIX THE DESIGN LB DB Cache Queue CDN LOCKED IN ✓

This is a system design question.

One server stops. The request finds another way.

Answer one a day.

Pick an answer. It takes one minute.

Read why, in two minutes.

See the path that kept the site up.

Keep a streak. Try a lab.

Come back each day. Fix a broken design in the lab.

// Today

Something new every day

A fact, a few news summaries, an interview tip, and a work lesson. New each day. No account needed.

// AI system design

The new question family

Interviewers now ask you to design ChatGPT, a RAG pipeline, or an agent that uses tools. They grade the numbers. A diagram without a latency budget and a token cost does not pass.

Time to first token

Keep it apart from the time between tokens. The first one waits for retrieval and the prefill. The rest stream.

GPU capacity for each tenant

A batch size, a queue, and a limit per tenant, so one customer cannot take every GPU.

Token cost per month

Users times requests times tokens times price. Then the levers: a cache, a smaller model, a shorter prompt.

AI system · One requestdesign chatgpt
One request through an AI system The prompt goes to retrieval, then to the model on a GPU. Tokens stream back one at a time. A timeline shows the time to first token and the cost of the request. Prompt Retrieve vector index Model 1 GPU · batch of 8 ANSWER Putacacheinfrontofthemodelforrepeatprompts. TTFT 380 ms 0 ms 1.4 s
// What you get

Six ways to practice

Each one takes a few minutes. Do one a day, or do them all.

One question a day

Pick an answer. Then read why it is right or wrong. Every interview topic, from caching to AI systems.

System Design Lab

A system is broken. Add the missing parts to the diagram to fix it. Then explain your fix to an AI reviewer.

Signal Tips

A short tip after each question. It tells you the key idea, the common mistake, and how to say it in an interview.

Streaks and XP

Answer one question and your streak lives another day. Earn XP for each correct answer and watch your level go up.

Coding Patterns

Read a coding problem and name the pattern that solves it: Sliding Window, Two Pointers, BFS, and more.

Daily news

A few articles each day on AI, architecture, security, and hiring. Each one comes with a short summary so you can decide what to read.

// System Design Lab

Fix a broken system

The Lab gives you a system that failed. You add parts to the diagram to repair it. Then an AI reviewer asks you why your fix works.

01

Read what went wrong

A short report tells you how the system failed

02

Repair the diagram

Add the parts that would have stopped the failure

03

Explain your fix

The AI reviewer asks one question about your choice

04

Get a score and notes

What you did well, what you missed, and the trade-off to remember

Lab · Fix a broken systemincident 14
A lab in three steps A diagram with a gap between the service and the database. A cache slides into the gap. The reviewer asks why, and the answer appears. API Service ? DB hot Cache hit 92% 01 READ THE REPORT · DB AT 100% CPU 02 REPAIR THE DIAGRAM 03 EXPLAIN YOUR FIX REVIEWER Why a cache, and not a bigger database? Reads repeat. The cache answers them. The DB rests.
Patterns · Name the patternproblem 8 of 10
Name the pattern A problem card: a sorted list, find two numbers that add up to a target. Four pattern names. The cursor picks Two Pointers and a check appears. PROBLEM A sorted list and a target number. Find the two numbers that add up to it. 2 7 11 15 21 target 23 Sliding Window Two Pointers Binary Search Hash Map Backtracking Right. Move both ends in.
// Coding Patterns

Spot the pattern before you code

Most coding interview problems use one of about 15 patterns. We show you a problem, you name the pattern. Do this often and the right approach shows up fast.

Sliding Window
Two Pointers
Binary Search
BFS & DFS
Hash Map
Dynamic Programming
Greedy
Monotonic Stack
Union-Find
Backtracking
// Beyond interviews

Help for the rest of the job

Interviews are one day. These three guides help with the other days. How to talk to your team. How to code in front of someone. How to hand in a project.

// Why one a day

A little every day beats a lot the night before.

Two minutes a day

One question takes about two minutes. That is enough to keep your streak and keep the ideas fresh.

Wrong answers come back

A question you get wrong comes back in a few days. A question you get right comes back less often. This is called spaced repetition.

Real interview topics

The topics are the ones interviewers ask about: caching, databases, queues, load balancing, and now AI systems.

You can see your progress

Your streak, your XP, and a strength score for each topic. You can see which topics need more work.

Review · Wrong answers come backcaching
How a question comes back Fourteen days in a row. A question answered wrong on day one comes back on day four. Answered right on day four, it comes back on day eleven. DAY 1 · CACHING Where does a cache go? Wrong. Back in 3 days. DAY 4 · CACHING Where does a cache go? Right. Back in 7 days. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 3 days 7 days Miss it, and it comes back soon. Get it, and it waits longer.
// FAQ

Common questions

What is system design?
System design is how you plan a large software system. You decide which parts it has, how the parts talk to each other, and what happens when one part fails. Topics include caching, databases, load balancing, and queues. Many companies ask about it in interviews for senior roles.
What is AI system design?
AI system design is the same kind of planning for a system with a language model inside. You design how a request reaches the model, how the answer streams back, how the system finds the right documents, and how you keep latency and cost under control. Interviewers now ask about it by name.
What is a system design primer?
A primer is a short guide to the basics. Our primer starts with scalability and ends with AI system design. It is free to read on this site.
How do I prepare for a system design interview?
Practice a little every day. On SD Primer you answer one question a day, read why the answer is right, and come back tomorrow. Questions you get wrong come back later. Read the primer first if the basics are new to you.
Is SD Primer free?
Yes, the free plan gives you daily questions, the primer, the question bank, and the trends posts. The Pro plan adds the System Design Lab, custom practice sets, and more. Pro is sold in the iOS app.
Who is SD Primer for?
Software engineers with a system design interview coming up. Also engineers who want to get better at design in their day job. You do not need to know anything about system design to start.

Start with one question.

It is free. You do not need an account. Also on iPhone.