Why most interview practice fails before it starts
You walk into an onsite cold because the practice you've done doesn't match the loop you're walking into. ChatGPT will play interviewer if you ask, but it doesn't know that Google L4 weights system design scoping in the first 10 minutes, or that Amazon's Bar Raiser is a different kind of question than the team rounds. Friend mocks soften feedback because no one wants to tell you your STAR collapsed in round three. Generic interview-prep books cover everything and target nothing.
The result is a candidate who has practiced "interviewing" but hasn't practiced this interview. They've drilled algorithms but never been pushed on complexity analysis under time pressure. They've rehearsed STAR stories but never had a follow-up question break the structure. They've thought about system design but never been forced to scope before drawing.
A good mock interview reproduces the exact loop you're about to walk into — the question style, the signal weighting, the follow-up patterns, the time pressure — and tells you precisely where you would have failed.
Built from what candidates actually report
The question bank is assembled from public interview reports — Glassdoor, Reddit, 一亩三分地 — mapped to the company and level the question came from, with the source shown next to each one. Around that sits what those same reports say about how a round is run: how long it lasts, what the interviewer pushes on, which follow-up tends to arrive when an answer is thin.
That is why a Google L4 mock here looks different from a Meta E4 mock. Same role title, different questions, different follow-up style. The AI loads the right set before the round starts. What it cannot do is read a company's internal scoring sheet, and nothing on this page claims otherwise.