Real interviews don't test recall.They test what survives"why?"
An AI examiner that presses every claim with "why?" — then "why?" again — until your logic holds or breaks. Train the way second rounds actually test you.
50 seats at $30/mo, locked for life.
State the definition of Set.
DEF · Mathematical Foundations · ~3 min · 2 hints available
Give me the formal definition of a set. Be precise about membership and what makes a collection well-defined.
A set is a collection of distinct objects, called elements. We write x ∈ S to say x is a member of S.
You have membership and distinctness. Now what makes the collection itself well-defined? Consider R = {x : x ∉ x} — what happens?
Prep for the interviews that decide your career
You didn’t fail because the problem was hard.You failed because it wasn’t familiar.
One change—and your playbook breaks.You stall. You guess. Time runs out.
300 reps made you fast.None made you adaptable.
Define. Understand. Apply.
The examiner decides when each layer is proven. You don’t move on until it is proven.
- DEF
Can you actually define it?
State it precisely. Then defend every word.
ExaminerDefine a limit of a sequence.
YouA sequence aₙ converges to L if for every ε > 0 there exists N such that |aₙ − L| < ε for all n ≥ N.
ExaminerWhat does "for every ε > 0" actually mean? Why not just "small ε"?
- UND
Can you explain why it works?
Explain why it works. Derive, don't recall.
ExaminerWhy does the chain rule work? Don't compute—explain.
YouBecause the derivative is a local linear approximation. If f locally scales by f′(g(x)) and g locally scales by g′(x), composing them scales by the product. The chain rule is just "linear approximations compose by multiplication."
ExaminerRight—now where does that argument actually fail?
- PRB
Can you reason through it under pressure?
Solve under pressure. Every step justified.
ExaminerA drunk man takes n steps on a line, each ±1 with equal probability. What's his expected squared distance from the origin?
YouLet Sₙ = X₁ + ⋯ + Xₙ where each Xᵢ is ±1 with prob ½. Then E[Xᵢ] = 0 and E[Xᵢ²] = 1. So E[Sₙ²] = E[(ΣXᵢ)²] = Σᵢ E[Xᵢ²] + Σᵢ≠ⱼ E[XᵢXⱼ] = n + 0 = n.
ExaminerYou dropped the cross terms. Justify it.
Get tested first. Fix what breaks.
The standard pattern is study-then-test. We invert it. The mock goes first; the curriculum follows what it exposes.
- 01
Get stress-tested first
A 30-min mock opens the loop. No scaffolding. Pressure exposes what study hides.
- 02
See where you broke
Post-session review names the topics — and the layer — your reasoning broke at.
- 03
Go deeper on weak areas
Examiner sends you to the layers you need to rebuild. Targeted, not blanket.
- 04
Repeat until mastery
Re-enter the mock. Watch the gaps close. Examiner decides when you’ve proven it.
Thirty minutes. Real pressure.
A timed session that mirrors the real interview. One to three problems, no scaffolding — the examiner stays in interviewer mode the whole way.
The clock starts immediately. No pause, no restart.
Ends in a ruling, on the record. The debrief opens from there.
Paced by the examiner, not a script.
Proven or broken — and exactly where.
Every session ends in a verdict. The examiner shows whether your own call matched its ruling, then names the criterion your reasoning failed at. No participation credit.
Excellent composure; near hire bar.
The moment the verdict was decided
Question 1 — the first question that was not defended. The debrief opens its turn-by-turn from here.
377 problems. Drill any of them live.
Browse the catalog by pillar, layer, or interview style. Every problem can be run live against the AI examiner — graded against a rubric, not an answer key.
Best-of-Seven Series
Google · Probability
Runs live · rubric-graded
Unfair Coin Detection
Probability fundamentals
Runs live · rubric-graded
Russian Roulette Strategy
Brainteasers
Runs live · rubric-graded
Why question banks fail.
Shows you the correct solution. You read it and move on.
Asks you why you got it wrong. You prove you understand the gap.
You decide. You move on when you feel ready.
The examiner decides. You move on when you've proven mastery.
Pattern recognition. Recognize the type, recall the solution.
First-principles reasoning. Derive under pressure, from nothing.
Memorized solutions collapse when one variable changes.
Fundamentals hold. Variations don’t require new machinery.
You’ve seen a lot of problems. You freeze on the one you haven’t.
You own the fundamentals. Variations stop surprising you.
Nothing here is improvised.
Every problem, every checkpoint, every follow-up the examiner can ask is hand-written by the founder and verified by experts before it ships. Security gates check every reply before it reaches you — the AI is never allowed to improvise.
- 1
Hand-authored.
The AI does not write math. It does not generate questions. Every problem, checkpoint, and follow-up is written by the founder. There is nothing for the model to make up on its own.
- 2
Expert-verified.
Every problem and every rubric step is reviewed by subject-matter experts before it goes live.
- 3
Locked.
The verified set becomes a locked library — approved question shapes, locked rubrics. The examiner grades against that rubric, never against its own opinion.
- 4
Gated at runtime.
Every reply is checked before it reaches you: did the AI pick from an approved shape? Did it stay on-rubric? If anything is off, the reply is dropped.
- OutputReaches you
“Why log? Why specifically that function?”
Round three at WorldQuant. I’d spent four months preparing for it.
When he asked me to explain logistic regression, I relaxed a little. Finally, something familiar.
I started answering.
Then he cut in:
“Why the log?”
And my brain just… froze.
Not because I hadn’t used logistic regression before — I had, a lot. But I’d only learned how to explain it well enough to pass interviews, not deeply enough to defend it.
I remember sitting there in silence, trying to think of anything to say.
Nothing came.
I knew right then the interview was over.
That moment changed how I think about interview prep. Most people practice polished answers. Great interviewers look for the cracks underneath them.
That’s why I built the thing I wish I’d had back then.
Founding access, while the cohort is open.
The first 50 members lock the founding rate for life. Once the cohort fills, the price returns to standard — and the founding rate never comes back.
Lifetime
One payment. Full access for the life of the product.
- Everything in Founding
- No subscription, no renewals
- Every future pillar included
- Founding-member perks
Founding
A third of the standard price — locked for as long as you stay.
- Full curriculum — all four pillars
- Unlimited adversarial sessions + timed mocks
- Mastery tracking across sessions
- Rate locked for life — direct line to the founder
Enterprise
For teams, bootcamps, and universities prepping candidates at scale.
- Seats for your whole team or cohort
- Centralized billing & onboarding
- Custom problem sets per role
- Dedicated support from the founder
Asked before buying.
Candidates preparing for rigorous technical interviews — quant trading, ML, research, math grad school. Self-learners who can recognize a problem but stall when it’s rotated 30°.
The real interview won't go easy on you.Neither will this.
Sit a full mock now and find the gaps while the stakes are zero.
Cancel anytime.