System design interview practice

A coding problem has a right answer. System design does not: it is an open-ended conversation about trade-offs, and about how you handle ambiguity when there is no single correct diagram. Practicing that alone is the hard part. A book can explain consistency models. It will not ask a follow-up when your answer leaves a gap.

There are two ways to work on it here, and it is worth knowing which is which.

• Course exercises. The free System Design course runs 224 lessons. Each one teaches a building block, then asks you to write your own design answer before it will show you a model one. Nothing is auto-graded on purpose: you save your answer, unlock the model answer, and compare. That is free, it needs only an account, and it never spends an interview session.
• Drills. 12 open-ended briefs, each the whole round with an AI interviewer, ending in a score and a written report. A drill spends one of your monthly interview sessions, 8 of which come with the free plan. You start them from the Drills section of the course page.

1. You pick a brief from the drill list. They run from a URL shortener at the easy end through a rate limiter, a distributed cache, and a notification system, up to Instagram, Uber, and a news feed.
2. Some briefs ask which company you are interviewing at, because the expectations differ. Then you press start.
3. The round opens in three columns: the brief with its functional and non-functional requirements and constraints, a notes editor, and the interviewer.
4. The notes editor is not blank. It opens on a skeleton with four headings to fill in: requirements clarification, high-level architecture, data model, and API design. You write prose and bullets, not a diagram. There is no whiteboard here, deliberately, and your design lives as text.
5. The interviewer does not wait to be asked. It reads what you are writing and interjects, pushing on choices you have made and on the ones you have skipped. You answer by typing or by voice.
6. When you are ready you press Submit Design. The clock counts up rather than down, and nothing is submitted for you, so a round ends when you end it. The exception is a company with a published time limit, which gets a real countdown.

Course exercises are narrower than a full brief, which is the point: they drill one move at a time. This one is from the first level, on clarifying a vague prompt.

Write the 3 to 5 clarifying questions you would ask for the bare prompt “Design Twitter”, and for each one show how a likely answer narrows the design.

Alongside it sits a short “think about” list rather than a hint that gives the move away: which product slice is actually in scope and what you will explicitly defer, what you need to know about actors, scale, and read/write mix before drawing anything, and how you avoid analysis paralysis inside three to five questions.

A drill brief is the other end of the range: an open statement of a product to design, with its requirements and constraints listed, and 45 to 60 minutes of conversation to get there.

Submitting produces a score out of 100, broken into four parts. The interviewer reads the conversation and your design notes together, so a strong document with no explanation and a strong explanation with an empty document both score badly.

Communication carrying the same weight as architecture is not padding. In a real design round the interviewer only ever sees the version of your design that you managed to say out loud. Treat the number as a rough band rather than a verdict, though. The written report below it is the part worth reading twice.

After a drill, the report comes back in four blocks:

• The short version. One paragraph on how the round went overall.
• What worked. The specific moves worth repeating, quoted back from what you actually did.
• Fix next. The gaps, ordered, rather than a list of everything that was not perfect.
• An action plan. What to do before the next round.

The interviewer also closes out in the chat with a two or three sentence debrief, so the round ends like a conversation rather than a page refresh. The score and breakdown are saved to your session history.

Course exercises end differently, because nothing there is graded. You save your answer, the model answer unlocks, and you compare the two yourself. Some exercises add a short rubric you score yourself against before the model answer appears, in that order on purpose: reading the model answer first makes an honest self-assessment impossible. Those self-scores stay in your browser and are never sent anywhere.

If you want to see written feedback before spending anything, the graded sample sessions are readable end to end.