It reads the board, not a checkbox
Your design is a typed graph, so the marking is graph questions: is there a cache on the read path, does the client reach the store without passing a service, is the write path async.
Devscape — system design interview practice
Take a real interview question, draw the architecture yourself, and have it graded on the spot — so you find your weak spots here, not in the interview.
No signup required. Open a challenge and start designing in ten seconds.
27 challenges215 architecture checksInstant feedback
Practice
27 system design challenges, each one carrying the checks an interviewer actually pushes on — 215 of them in all. Draw your answer, submit it, and get a score, the misses, and the follow-up question each miss would have earned you. Then fix it and go again.
No signup required. 8 beginner · 8 intermediate · 11 advanced · graded in your browser, signed out, as many times as you like.
Your design is a typed graph, so the marking is graph questions: is there a cache on the read path, does the client reach the store without passing a service, is the write path async.
Nothing comes back as “incorrect”. It comes back as the question an interviewer would have put to you, with the answer they were listening for underneath it.
Capacity estimates are graded by order of magnitude — within 3× either way, because being out by 40× is the mistake worth catching and nobody cares about 15%.
Every question ships with the design an interviewer had in mind, compared to yours box by box and link by link. Differing from it is allowed; not knowing you differ isn’t.
The difference
Excalidraw, Lucidchart, a whiteboard, a napkin — all of them let you draw a system. None of them tell you whether the system works.
| Aspect | Diagramming tools | Devscape |
|---|---|---|
| What you do | Draw a diagram | Design a system against a brief |
| When you finish | Nothing happens | It is marked, in seconds |
| Who says it is right | You do | A rubric that reads your graph |
| What a mistake looks like | You never find out | The question an interviewer would ask |
| The numbers | Not part of it | Capacity estimates graded too |
| What you leave with | A picture | A score, and a list of what to fix |
You can still just draw — the editor below is the whole of one, and it is free. It is simply not the reason to be here.
You have built services. You have not had to defend a whole system out loud, against numbers, in forty-five minutes.
You know the answer. This tells you whether the design you actually drew contains it — which is the thing being marked.
SDE II, staff, and every rung where the round stops being about code and starts being about trade-offs and capacity.
Years out from the last interview. Start on a pastebin, work up to fan-out at Twitter scale, at your own pace.
Twelve seconds
The palette
6 to start with, and none of them sacred. Name your own, give it a colour, drag it where it belongs — and the palette belongs to that board alone, not to the app.
The links
You say what connects to what. Where the line goes, and what it steps around on the way, is not your problem.
Boards
Name them, switch between them, rename or delete them — most recently touched at the top. Each keeps its own diagram and its own palette, so none of them has to agree with the others.
Sharing & export
A link anyone can open, read only, switched off whenever you like. Or a file — and the file is drawn from the board, not scraped off the screen.
Half a second after you stop.
Sign in later; your boards come with you.
Ctrl+Z for a box, a toast for a board.
Paste the JSON, or pick the file.
Fork any board someone shares with you.
One click away, and you see the reply.
Shortcuts
A board is stamped, not dropped.
Private until you say so. Revocable after.
They get the picture, never the editor.
JSON out, JSON back in.
By designing systems, not by reading about them. Pick a challenge, read the brief and the numbers it comes with, draw the architecture on the board — services, databases, queues, caches and the links between them — do the capacity arithmetic, and submit. You get a score and a list of what an interviewer would have pushed on. The loop is worth repeating: attempt, read the misses, redraw, submit again. Passive reading is what leaves people stuck, because an interview tests whether you can produce a design under time, not whether you have seen one.
All 27 of them are listed on the practice page and linked from the homepage. They run from a first pastebin, a URL shortener and a rate limiter up to Twitter’s timeline fan-out, WhatsApp, Uber, YouTube, Google Docs, a payment system and an LLM inference API — sorted beginner, intermediate and advanced, with the checks each one is marked against shown on the card.
Your board is a typed graph, so most of what an interviewer pushes on is a question about it: is there a cache on the read path, does the client reach the store without passing through a service, is the write path async, is that capacity estimate the right order of magnitude. Those checks are run over your design in the browser the moment you submit. The same design always scores the same, and anything that cannot be checked is left out of the score rather than guessed at.
Yes — no account is needed and none is asked for. Open a challenge, draw the design, submit it and read the feedback, all of it signed out and with no limit on attempts. Signing in exists so your boards follow you to another browser, not to unlock anything.
Yes. There is nothing to buy, no per-attempt cost and no trial. The grading runs on your machine, which is most of the reason it can be.
A score out of 100, every check that passed with the evidence from your board, every check that failed written as the question an interviewer would have asked, the answer they were listening for, and a comparison against the model answer — which boxes and links it has that yours does not.
You can, and nothing there will tell you the design is wrong. A drawing tool takes whatever you put on the canvas; it has no idea what the boxes mean, so it cannot notice that every timeline read goes straight to the database or that your write estimate is out by 40×. Here the board is a typed graph and the question comes with a rubric, so finishing the drawing is the point at which the work starts rather than the point at which it stops.
An article tells you the answer. Here you draw one first and then find out what you left out, which is the part an interview actually tests. The brief comes with real numbers, the capacity questions are marked, and the follow-ups arrive after you have committed to a design rather than before.
Contact
A question about the practice questions, something about your account, or anything the FAQ above didn’t answer. It goes to a person, not a queue — there is nobody else here.
Find out what you’d have missed while it still costs you nothing.
No signup required. The account is for carrying your boards between browsers, and nothing else is behind it.