Answer one moment of the interview at a time: scope the problem, do the math, survive the deep dive. Every answer is scored on the five signals interviewers write up, calibrated to E4, E5 and E6. Then you see how a production open-source system solved the same problem, pinned to the commit.
No 45-minute commitment. A drill is one part of the interview, the part where candidates actually lose the offer.
Each question splits into five drills: scoping the requirements, a back-of-the-envelope estimate, and three deep dives the interviewer would push on.
A score on each signal from very weak to very strong, the key points you hit and the ones you missed, and the one thing to fix first.
The model answer, then the real system: which table, which lock, which queue a production codebase uses for exactly this problem.
Two users tap the same seat within 50 ms of each other. You have Postgres and Redis. Walk through exactly what stops a double sale, and what each user sees.
With a free account, the interviewer doesn't stop at your first answer. It reads what you wrote and asks the follow-up a real interviewer would ask next, up to three deep. That chain is where hand-wavy answers stop working.
System design interviews aren't graded on whether the boxes look right. Interviewers write feedback against a handful of signals, and the hiring committee reads the signals, not the diagram. Revibe grades every answer on the same five:
very weak → weak → strong → very strong · strong means the hiring bar at E5
Each question also spells out what E4, E5 and E6 look like on that specific problem. On the ticketing question, E4 solves double booking once the interviewer points at it. E5 raises seat contention without being asked. E6 argues fairness against throughput from the business side.
Three questions, fifteen drills. A short list on purpose: each one comes with a hidden spec, a follow-up tree, a level rubric and a real codebase to check your answer against. More are on the way.
The RAG question is the new kid in system design loops: an LLM assistant over a company's wikis, drives and tickets, where the hard part isn't the embeddings. It's never retrieving a document the asker isn't allowed to open, and forgetting a document the moment it's deleted. If that's your loop, the RAG architecture hub breaks down seven real RAG codebases.
The standard prep answer for "design Ticketmaster" is a diagram someone drew for a video. It has never sold a ticket.
Every Revibe question is paired with an open-source system that solves the same problem in production: pretix for ticketing, Novu for notifications, Onyx for enterprise RAG. After you answer, the reveal points at the exact code that handles that drill, pinned to a commit so the line numbers don't drift.
Reading a model answer feels like learning. Producing one under a follow-up is the actual skill. The drills make you do the second before showing you the first.
The reveals come from the same engine behind Revibe's architecture diagram generator and the gallery of real codebase breakdowns, which stay free to read when you want to study a system end to end.
Yes. Every question, prompt and drill is free to read with no account. You get one graded drill a day without signing up. A free account raises that to 30 graded drills a day, unlocks the interviewer's follow-up questions, and saves your scores into a skill map. No card required.
A language model grades your answer against a written rubric and a hidden spec listing the key points a strong answer covers for that drill. You get a level on each of the five signals plus the points you hit and missed. Treat it as a demanding sparring partner, not a hiring committee: it is consistent and available at 2 a.m., and it does not know your recruiter.
E4, E5 and E6, roughly mid-level, senior and staff. Each question spells out what a passing answer looks like at each level, and the grading says where your answer landed. Strong on a signal means the hiring bar at E5.
The ticketing system question shows up in Meta and Amazon loops. The enterprise RAG question has been reported in an Amazon Applied Scientist technical screen. The notification system is a staple across consumer-app companies because it tests fan-out and idempotency in one prompt.
Yes. The enterprise RAG question covers scoping an everything-at-once prompt, sizing the index, permission filtering inside retrieval, keeping the index fresh as documents change, and tuning and evaluating hybrid retrieval, then shows how Onyx handles each one.
Books and courses show you a finished answer. Drills make you write one, grade it, push on it with a follow-up, and only then show you the answer, from production code rather than a whiteboard. They pair well: read for the patterns, drill to find out whether you can use them.
No. A full mock with a human is still the best rehearsal for the real 45 minutes. Drills are the reps between mocks, aimed at the specific moments where answers fall apart: the estimate nobody did, the race condition nobody mentioned, the follow-up nobody expected.
One drill. Graded on the interview rubric. No account needed for the first one.