Atlassian Karat Interview Questions: Format, Patterns, and How the Screen Works

Atlassian Karat interview questions come from a screened technical interview, not from a teammate on the hiring panel. Karat is a third-party interviewer. You code for about an hour, they write a scorecard, and Atlassian decides whether you move on. Treating it like a casual phone chat, or like a silent online assessment, is how prepared people still fail it.

InterviewTruth does not have a dedicated Atlassian question page. This guide is the format and the patterns. For reports candidates are posting across companies, use recent questions. Karat's exact prompts rotate and are not ours to republish as a leaked set.

How the Atlassian Karat round is structured

Candidates describe a single session of about 60 minutes with a Karat interviewer in a shared editor.

Block	What it is for
First few minutes	Intro, confirm the language, confirm you can see the editor
First problem	A practical coding problem you are expected to finish and test
Second problem	Harder, only if the first one is in good shape
Last few minutes	Your questions, if any time is left

The interviewer is not your future manager. They are there to watch you solve, to ask what your code does, and to record whether it worked. Being warm is fine. Spending ten minutes on small talk is not, because the second problem disappears when the first one runs long.

A common failure is the reverse: rushing a buggy first solution to "get to the hard one." The scorecard cares that the code you submitted actually works. A clean first problem and a partial second problem beats two broken drafts.

This is a different test from Goldman Sachs OA questions, which are timed and unattended. Karat is live. Talk. If you go silent for several minutes, the interviewer cannot score a thought they did not hear, and they will not invent one for you.

What the screen is scoring

You will not see the rubric. You can still aim at the things a coding screen has to record:

Did you understand the problem before writing code, including an input you asked about?
Does the code work on the example and on a case you added?
Is the complexity honest? You should say it, including the sort if you sorted.
Can someone else follow it? Names, a short comment where the invariant is non-obvious, no abandoned half-algorithms in the file.
Did you test? Run the example. Then an empty input, a duplicate, or a one-element input. Fix what breaks.

Karat interviewers will nudge. A nudge is not an insult and it is not a hint to ignore. If they ask "what happens on an empty list?", answer that question in the code, not with "the problem didn't say."

Atlassian Karat interview questions by type

These are the families engineering screens at this stage resolve to. They are practice targets, not a claim that any specific wording is on your link.

Hash maps, arrays, and counting

Practice shape: given a list of events with user ids and timestamps, return the users who performed an action at least k times inside any window of length t.

That is a hash map of queues or a sliding window. The product noun (pages, tickets, comments) does not matter. Say the structure, then code it.

Also: two-sum and its variants, group by a key, first non-repeating element, merge intervals.

Strings and light parsing

Validate a pattern, tokenize a simple path, or find a window that contains every required character. Ask whether the alphabet is lowercase only. If it is, a fixed array is easier to get right than a map you update in three places.

Trees and graphs you can finish

BFS on a grid, level order, "is there a path," and a tree traversal that accumulates an answer. Karat problems in this family are usually direct. The bugs are visited-sets and null children, which is exactly what a rushed candidate ships.

Implement BFS often enough that the queue and the visited set appear without a search through your memory.

A small class or API

Some sessions ask you to implement a type with a few methods: a counter, a logger with retention, a structure that supports add and lookup. This is the closest the screen gets to design.

Do this first:

Repeat each method and its complexity
Name the fields
Then code
Test two calls in a row, because the second call is where stale state shows up

If you want a longer version of that habit, how to crack LLD interviews is written for the onsite, and the first half applies here.

What to skip

Segment trees, obscure DP, and system design diagrams. The Karat hour is not the Atlassian system design round. If you spend it drawing services, you will not have working code.

How to talk during the hour

Use a script you can follow while nervous:

"I'll restate it as: given X, return Y."
"I'm assuming input can be empty and that ids fit in memory. Stop me if that's wrong."
"Brute force is nested loops, too slow if N is large. I'll use a hash map, O(n) time and O(n) memory."
Code the happy path, then the edge you named.
"I'll run the sample. Expected output is this. Actual output is this."
If they offer a second problem, leave the first file in a state you would not be ashamed to reread.

If you are stuck, say what you know. "I need the count of items in the window as it slides, and I don't want to rescan" is a solvable sentence. Silence is not.

Pick the language you can test without thinking. Python is a common choice because the problems are short. Java is fine if your Python is the thing you would be debugging at minute 40.

After the scorecard

A pass is not an offer. Atlassian still runs its own interviews, often including:

A values conversation. Read Atlassian's current published values before that round and bring stories that match how you actually work. Do not recite a list.
A system design or project deep-dive for experienced candidates
Team conversations

The Karat screen will not ask you to perform the values round. Do not skip it later. Behavioral structure from effective storytelling in hiring manager interviews transfers.

If Karat does not pass you, the useful review is mechanical: did the first problem actually run, did you test an edge, did you talk. Those three are practiceable in a week. Collecting more "Atlassian-tagged" problems is not, by itself, a review.

A one-week plan

Days 1–2. Six hash map and array problems, 30 minutes each, in a shared-editor mindset: talk to an empty room. Run every sample.

Days 3–4. Four BFS or tree problems and two "implement this class" problems. On the class problems, write two sequential calls as tests before you declare victory.

Day 5. One 55-minute mock with two problems. Stop at 55. The point is to feel the second problem slipping away and to practice saving a correct first solution anyway.

Day 6. Redo only the problems you could not finish. Fix the specific bug (visited set, off-by-one, stale field), do not re-read the editorial and nod.

Day 7. Rest the syntax. Glance at recent questions for anything filed under a coding screen, and do not start a new topic that evening.

Frequently asked questions

Who interviews you in the Atlassian Karat round?

A Karat interviewer, not an Atlassian engineer. They write up the session. Later rounds, if you advance, are with Atlassian.

How many Atlassian Karat interview questions are there?

Usually one required problem and a second if time allows, inside a single hour. Finishing the first correctly matters more than touching the second.

Is the Karat round the same as the online assessment?

No. Some processes also have an unattended assessment. The Karat round is live coding with a person. If your email says Karat, prepare to talk. The Goldman Sachs OA guide is the unattended version of a similar coding level, at a different company.

What language is best?

The language you can debug in public. Changing to a fashionable language the week of the screen is a common own goal.

Can I use the internet or an AI assistant?

Follow the instructions the interviewer gives you at the start. Do not assume an assistant is allowed the way some 2026 online assessments allow one. If you are unsure, ask in the first minute. Asking is free. Guessing is not.

Final thoughts

Atlassian Karat interview questions are a one-hour live coding screen: a practical first problem you test, a harder second problem if it is earned, and a scorecard written by someone who is not on the team. Talk through the pattern, run the sample, and fix the edge they ask about.

Practice that for a week. Then skim recent questions so you are reading current reports instead of a stale prompt list. Good luck.