The real round, before the real one

1. 01 of 04

   System design

   25 to 45 minWhiteboard it or load-test itProve it

   Pick a problem and talk through the design with an AI interviewer. Draw it on a canvas. At the scalability step, Prove it runs a real load test on your own design.

   • 14 problems, including Payment System (UPI and Cards) and Ticket Booking System (Movies and Trains).
   • 5 steps with a time guide: requirements, API design, high-level design, deep dives, wrap-up.
   • 8 problems have Prove it. You say what your design can hold, then the test says if you were right.
   • Or load-test it: build your design in the simulator and the interviewer tests your claims on the real engine (one problem so far).

   Interview · Design a URL ShortenerSample

   Requirements up to 8 minAPI design up to 8 minHigh-level design up to 20 minDeep dives up to 15 minWrap-up and score up to 5 min

   InterviewerGood. What happens to those 240 reads per second if the cache restarts during the evening peak?

   YouAll 1,200 go to the database for a while. I would warm the hot keys first and cap the database.

   Prove it10x growth (12,500 redirects/s)Heldp95 310 ms · Database at 92%

2. 02 of 04

   Machine coding

   75 min, 90 at mostPython, Java, JavaScript, C++Hidden tests

   A low-level design round. Clarify the problem, design the classes, code a working core, then handle a new requirement the interviewer adds.

   • 10 problems: Parking Lot, Splitwise, Elevator, Rate Limiter and more.
   • Run is private. Submit runs hidden tests on your real code, and the interviewer reacts in words.
   • The new requirement always arrives in the same words, so the tests match what you were told.

   Machine coding · Parking LotSample

   Clarify up to 10 minClass design up to 15 minCode up to 35 minNew requirement up to 15 minWrap-up up to 5 min

   Your class: ParkingLot

   park()slotOf()unpark()availableSlots()parkedPlates()totalRevenue()

   Interviewer, after you press SubmitI tried a 10 hour stay and the fee looked too high.

   NextNew requirement: monthly passes and overflow parking

3. 03 of 04

   Resume grilling

   About 35 minYour own projects

   Paste your resume. The interviewer picks the two points it can check best, and digs in: what you did, how it works, the numbers, what broke.

   • Your resume is cleaned first. Emails, phone numbers and links are removed before anything is used.
   • The interviewer notices numbers that do not add up, and asks about them without accusing.
   • You get a verdict for each point, the red flags with your exact words, and a rewrite of the weak line.

   Resume grilling · Your two pointsSample

   Your two pointsFirst project up to 12 minSecond project up to 10 minMake it 10x up to 8 minWrap-up up to 3 min

   1Cut API latency by 40% (800 ms to 320 ms) by adding a Redis cache.

   2Moved the nightly report job from cron to a queue with 4 workers.

   InterviewerYou wrote 40%, but 800 ms to 320 ms is a 60% drop. Which number did you measure, and how?

   Email removedPhone removedLinks removed

4. 04 of 04

   Incident

   12 to 15 min8 investigationsA clock is running

   Production is down and customers are affected. Read the dashboards, logs, recent changes and traces, stop the damage, and name the root cause.

   • Every panel you open costs minutes, and the number of affected customers keeps growing until you stop it.
   • You choose from real actions: roll back, scale out, fail over, rate-limit, tell people.
   • The score is worked out by code from what you did. The AI cannot change it.

   Incident · Customers do not get their login codesSample

   Alert

   OTP login success 41%, SMS queue over 100,000 and growing

   Started 19:09, you were paged at 19:12

   Dashboards1 minLogs2 minRecent changes1 minTraces2 minRunbook1 min

   Customers affected now7,200

   More every minute+2,400

AlsoA coding interview: one problem, about 45 minutes, in the language you choose, run against a real code judge.