AI Interviews
The interview has an AI in it now. Practise the new one.
A real codebase, hidden tests you cannot read, and a clock. The assistant is allowed, and how you steer it is the thing being marked. Three rounds, and a debrief afterwards that quotes your own session back to you.
₹99 for one · no subscription · one credit opens any of the three
Round one
AI coding interview
45 to 55 minutes · four phases
A small codebase that already works, mostly. Read it, fix what is planted in it, add what was asked for, then make it survive an input a hundred times bigger.
- 51 problems in Python, TypeScript and Java
- 3,152 hidden tests you never see
Round two
Code review
20 to 30 minutes · one pull request
A teammate opens a pull request and asks you to look before it ships. Defects are hidden in it, and so is code that looks wrong and is perfectly fine.
- 120 pull requests across fourteen areas
- Graded as a junior, a mid-level or a senior
Round three
Low-level design
45 minutes · one document
No code in this one. You are given a system to design and an interviewer who reads what you wrote, asks about the parts you left vague, and moves the requirement before the clock runs out.
- 25 systems, from a hospital lift bank to a payment router
- The first one is free, and coached beat by beat
Or bring friends and earn one free
Your friend gets a free code review. When they take it, you get one too, and every fifth friend earns you a full AI coding interview.
Why this round is different
The assistant is not banned. It is the exam.
Interviews used to test whether you could write the code alone. Now the laptop has an assistant on it and everyone knows it, so the question has moved: can you say what you want, judge what comes back, and stand behind it. That is the part nobody has practised, because there has been nowhere to practise it.
41def balance(self, account):42 total = 043 for e in self._entries:44 total += e.amount45 return totalEvery entry, every call. Fine for ten. Not for two hundred thousand.
Assistant
Keep a running balance per account. Do not touch the entry list.
A dict keyed by account, updated where the entry is appended.
You are marked on how you steer it
Prompts that name the class and the approach you want score well. Pasting the problem in and taking whatever comes back scores badly, even when the tests go green.
Hidden tests settle it, not an opinion
Every phase has tests you never see. The last one has a time budget that the obvious answer misses, so "make it fast" is a real instruction rather than a suggestion.
The debrief quotes you
Not a score out of ten. Your own prompts and answers, read back with what they showed, and three things to practise next.
Inside the rounds
What the clock is actually for.
The two coding rounds come in Python, TypeScript and Java, with the same codebase, the same defects and the same tests in each. The third has no code in it at all. Here is what each one asks of you, minute by minute.
Round one
AI coding interview
You are handed a small codebase that already works, mostly. Four phases, one clock, and the assistant beside you the whole way.
- 1
Read it 10 min
Three questions about how the code works, answered in your own words. No running it, no assistant doing it for you.
- 2
Fix it 10 min
Three defects are planted in it. The tests that ship with the code pass; the hidden ones do not.
- 3
Extend it 15 min
Add what the product asked for, without breaking what already worked.
- 4
Make it fast 15 min
The input grows by a few orders of magnitude and the obvious version stops finishing. Find the structure it wanted.
51 problems, 3,152 hidden tests, graded on approach, control over the assistant, and whether you ran the code.
Browse the problemsRound two
Code review
A teammate opens a pull request and asks you to look before it ships. Defects are hidden in it, and so is code that looks wrong and is fine.
Comment where you would on a real PR
Pick a line or drag a range in the gutter. Say what is wrong, why it matters, and what you would do instead.
Graded against the bar you pick
Junior, mid-level or senior. The same pull request, weighed differently: what you were expected to catch, and how much a wasted comment costs.
Approving can be the right answer
Some of these are nearly clean. Inventing a problem to fill the space costs you more than saying it looks fine.
120 pull requests across fourteen areas, scored on what you found, how accurate you were, how clearly you wrote, and whether the serious things got your attention first.
Browse the pull requestsRound three
Low-level design
No editor, no tests, no assistant. One document and an interviewer who has read it. Five beats on a single clock, and you can go back to an earlier one when the walkthrough shows you something, which is what designing actually looks like.
- 1
Scope 6 min
What you are building, and what you are refusing to build. You are held to that line for the rest of the hour.
- 2
Pieces 7 min
The objects and the lines between them. Two things that behave differently are two objects, however alike they look.
- 3
Shape 12 min
The classes: what each one owns, and the handful of methods it offers. One job each.
- 4
Walkthrough 12 min
One real request, followed through the classes you just wrote. This is where a design holds together or quietly does not.
- 5
Pressure 8 min
The requirement moves. Saying plainly that a piece would have to be rebuilt beats pretending it would not.
25 systems, graded at the bar you pick on judgement, modelling, what each object ends up owning, whether the thing holds together, and what happened when the requirement moved. Close a beat and the interviewer pushes back on what you actually wrote, so no two rounds on the same system go the same way.
Never sat one? The first is a walkthrough on a board everybody already knows, with a worked example of each beat beside you as you write. It is free and it does not touch your credits.
Browse the systemsPrice
Pay for the interviews you sit.
Every session runs a real sandbox and an AI grader, so these are counted one at a time rather than folded into a subscription. Buy one to see what the round is like. Buy five if you are in the middle of a hunt.
- A credit opens any of the three rounds
- Valid 30 days from purchase
- Nothing recurring, and no plan needed to sit one
Before you start
The things worth knowing first.
- How long does one take?
- Forty-five to fifty-five minutes for the coding round, twenty to thirty for a review, forty-five for a design. One clock runs across the whole thing. Closing the tab does not stop it, and reopening puts you back where you were.
- Which language?
- Python, TypeScript or Java, chosen when you start. The codebase, the defects and the tests are the same in each, so pick the one you would actually interview in. The design round does not ask: you write about objects, not in a language.
- What does the assistant know?
- In the coding round it has no tools and sees only the file you have open, so what you tell it is what it works from. The design round has no assistant at all: the model on the other side is interviewing you, and it will not answer a question you were asked.
- Do I need the plan as well?
- No. Interview credits stand on their own, and the plan covers the courses, sheets and roadmap. Neither one includes the other.
- What if I run out of time?
- The session closes and gets graded on what is there. A half-finished round still produces a useful debrief, which is rather the point of practising.
- Is this a real company's interview?
- No. It is built to match the shape of the round, not any one employer's. Nothing you do here is sent anywhere.
The first time you do this should not be the real one.
₹99 to sit one, ₹80 each if you take five. Pick a problem, pick a language, and start the clock.
Got a gift or invite code? Redeem it here.