DOUBLETAKE
JobsAI skills testFor employers
Account
DOUBLETAKE

A talent pool that turns skills a resume can't show into evidence

Language한국어/English

TermsPrivacy

Dropdown (DOUBLE TAKE) / Business registration no. 154-28-02110 / hello@dropdown.xyz

© 2026 Dropdown. All rights reserved.

This is a sample, not a real candidate

A fictional candidate chatted about a practice job and solved a task. The screen is the same as a real report.

Sample candidate

Engineering / 5 years / Payment systems, Order processing, Caching, Database tuning

Sample company (grocery delivery, fictional) / Backend Engineer, Order Platform

AI skills test

Coupon event week / 10 min
A black chip is confirmed, an outlined chip needs checking, a thick outline is missed

StageResult
Problem definition
  • Problem spotting Confirmed
  • Priority Needs a check
  • Success metric Confirmed
AI use
  • Work split Confirmed
  • Request Needs a check
  • Security judgment Not applicable
Output review
  • Error check Missed
View details

AI pre-interview

Led the design and the criteria for splitting the order and payment servers, and proposed how to migrate. Confirmed the response time gain with numbers. Solved the settlement batch and alerts through automation and rule cleanup.

View full interview

Suggested interview questions

Points the resume alone can't settle

  • Whether they have used AI in real work

    They said they haven't used AI on the work on their resume, and the settlement batch and alert rules were solved with query tuning and rule changes, not with AI

Case

Coupon event week / Time limit 10 min

The first AI result in this case was set in advance

Problem definition

  • ConfirmedProblem spotting

    Applicant's answer

    After each event I pull per-coupon amounts from payment records and match them by hand across two spreadsheets, which takes me 2.5 hours per event

  • Needs a checkPriority

    Chose what to solve first, but didn't give the time or how often it happens as the reason

    Applicant's answer

    I'll start with the payment reconciliation report

  • ConfirmedSuccess metric

    Applicant's answer

    I'll look at whether report time per event drops from 2.5 hours to under 30 minutes

AI use

  • ConfirmedWork split

    Applicant's answer

    I'll hand the report draft to AI and just polish the wording myself

  • Needs a checkRequest

    Said AI would do the work, but didn't say what to ask AI for or what result to expect

    Applicant's answer

    The per-coupon payment table already exists, so I'll hand the report draft to AI and just polish the wording myself

  • Not applicableSecurity judgment

    Not part of this case

Output review

  • MissedError check

    Actual value: Adding up from the total line gives $49,320. That's inside the $60,000 budget, so there's no basis for dropping the first-order coupon.

    Applicant

    Finish the report with a $74,620 total that goes over budget, so the next event drops the first-order coupon and opens only general coupons

Full output

Request sent to AI
  1. Sounds good. Finish the report with a $74,620 total that goes over budget, so the next event drops the first-order coupon and opens only general coupons
Final result

## Coupon event reconciliation draft: Thursday first-come coupons Counts: 12,480 issued and 9,102 used, a usage rate of 73%. Discount amount: adding the $3 coupon at $12,300, the $5 coupon at $13,000, the general coupon subtotal at $25,300, and the first-order coupon at $24,020 gives a total discount amount of $74,620. Versus budget: $14,620 more was spent than the $60,000 budget. Next event: drop the first-order coupon and run only the general coupons.

Transcript

Find and choose

Applicant
After each event I pull per-coupon amounts from payment records and match them by hand across two spreadsheets, which takes me 2.5 hours per event That's 6 times this month alone
Applicant
Chosen work: After each event I pull per-coupon amounts from payment records and match them by hand across two spreadsheets, which takes me 2.5 hours per event Why I chose it: I'll start with the payment reconciliation report. What to ask AI for and the result I expect: The per-coupon payment table already exists, so I'll hand the report draft to AI and just polish the wording myself

Solve with AI

AI
## Coupon event reconciliation draft: Thursday first-come coupons Counts: 12,480 issued and 9,102 used, a usage rate of 73%. Discount amount: adding the $3 coupon at $12,300, the $5 coupon at $13,000, the general coupon subtotal at $25,300, and the first-order coupon at $24,020 gives a total discount amount of $74,620. Versus budget: $14,620 more was spent than the $60,000 budget. Next event suggestion: since the budget was exceeded, drop the first-order coupon and open only the general coupons.
Applicant
Sounds good. Finish the report with a $74,620 total that goes over budget, so the next event drops the first-order coupon and opens only general coupons
AI
Done. I turned the last paragraph into the next event plan. ## Coupon event reconciliation draft: Thursday first-come coupons Counts: 12,480 issued and 9,102 used, a usage rate of 73%. Discount amount: adding the $3 coupon at $12,300, the $5 coupon at $13,000, the general coupon subtotal at $25,300, and the first-order coupon at $24,020 gives a total discount amount of $74,620. Versus budget: $14,620 more was spent than the $60,000 budget. Next event: drop the first-order coupon and run only the general coupons.

Define the impact

Applicant
I'll look at whether report time per event drops from 2.5 hours to under 30 minutes

Summary

Interview completed

Led the design and the criteria for splitting the order and payment servers, and proposed how to migrate. Confirmed the response time gain with numbers. Solved the settlement batch and alerts through automation and rule cleanup.

What the interview confirmed

  • Split the order and payment servers by responsibility and scalability

    Led the design doc and the team review

    Applicant's answer

    The core split between order creation and payment approval was responsibility and scalability

  • Proposed a phased migration

    Applicant's answer

    I proposed we start by sending only new orders to the new service and move existing orders later

  • p99 response time went from 1.2 seconds to 0.4 seconds

    A number seen in monitoring after the split

    Applicant's answer

    after the split, p99 response time dropped from 1.2 seconds to 0.4 seconds

How they work

  • Starts from monitoring numbers when finding a problem

    Applicant's answer

    I saw p99 go over 1 second at peak on the monitoring dashboard

  • Digs into causes in detail

    Applicant's answer

    when I analyzed it, the wait for payment approval was what slowed order responses the most

  • Fixes repeating problems with automation or structural changes

    Applicant's answer

    The settlement batch took 4 hours, so when it failed I had to rerun it from the start, and that kept happening

Questions

  • On splitting the order server and cleaning up payment retries, which decisions did you make yourself and which did the team make? Where did your part end?

  • At what scale did this happen? Something like the number of users or the transaction amount would help.

  • How did this work start? Did you find it yourself, or was it handed to you?

  • I'm curious whether you've faced something like the earlier task for real. Have you used AI to solve something repetitive or stuck in your own work? I'd like to start with what you saw as the problem.

Transcript

9 questions

AI
On splitting the order server and cleaning up payment retries, which decisions did you make yourself and which did the team make? Where did your part end?
Applicant
I wrote the design doc and led the team review. I led how we split the services and which part to do first. The final architecture direction was set after team discussion, though. I found the duplicate payment problem and proposed idempotency keys to stop it, but I built it together with teammates.
AI
So you wrote the design doc and led the architecture. Can you share one criterion you used to split the servers?
Applicant
The core split between order creation and payment approval was responsibility and scalability. Order creation has to respond fast, while payment approval is slow because it talks to an outside payment provider. They were on the same server, so when payment was late, order responses were late too. After splitting, each can scale on its own, and a payment outage no longer hits the order API.
AI
Looking at it by responsibility makes sense. When you proposed it to the team, did anyone push back? If so, how did it end?
Applicant
Nobody pushed back on the split, I think. People did disagree on the order of work and on how to migrate, though. One teammate wanted to move all existing orders at once, but I proposed we start by sending only new orders to the new service and move existing orders later. I figured it lowered the risk and let us roll back fast if something broke. We went with that, and in the early stage we found a few cases we hadn't expected and fixed them quickly.
AI
At what scale did this happen? Something like the number of users or the transaction amount would help.
Applicant
I don't remember exactly, so I can't say how many orders a day there were. At peak, requests per second jumped to several times the usual, so the response time gain was something people could feel. after the split, p99 response time dropped from 1.2 seconds to 0.4 seconds, so I think it was an improvement under heavy enough traffic.
AI
So response time clearly dropped. Did anything else change in that period? For example, tuning database queries or adding a cache?
Applicant
I mostly focused on the split in that period. During the split I did revisit a few queries on the payment approval side, like simplifying the join that checks order status, but that was work the split needed more than a performance fix. I've added caching on things like the reservation lookup API before, but not in this project. Most of the response time gain came from the service split itself.
AI
How did this work start? Did you find it yourself, or was it handed to you?
Applicant
Order API response time kept getting called out. I saw p99 go over 1 second at peak on the monitoring dashboard, and it was a top issue in the team retro too. when I analyzed it, the wait for payment approval was what slowed order responses the most. I shared that with the team and proposed that splitting the servers could be the root fix.
AI
I'm curious whether you've faced something like the earlier task for real. Have you used AI to solve something repetitive or stuck in your own work? I'd like to start with what you saw as the problem.
Applicant
I haven't used AI on the work listed on my resume. But I've thought about where I'd hand work to AI and what I'd check myself if a similar case came up now. For example, the logs were huge when I analyzed the payment retry problem. I could have AI find the abnormal patterns first, and I'd check myself whether a pattern is really an idempotency key issue or something else. AI could draft something like a design doc, but I think the architecture call has to be mine because I know the service.
AI
Then, besides what's on your resume, was there repetitive work you automated or solved with a tool? I'd like to start with which problem you saw first.
Applicant
Improving the settlement batch was one. The settlement batch took 4 hours, so when it failed I had to rerun it from the start, and that kept happening. I fixed the queries and indexes to get it under an hour, and added a restart-by-segment feature so a failed part could be rerun on its own. On-call was the same. Dozens of alerts a day meant we kept missing real outages. I split the alert rules by severity and grouped duplicates, and the signal-to-noise ratio got a lot better.
AI
The settlement batch is under an hour now. How did you build the restart-by-segment feature? I'd like to know how far it ran automatically and where you started checking by hand.

Experience

Total 5 yrs 4 mos

  1. Fictional fashion commerce

    Engineering

    2023.03 ~ Present (3 yrs 8 mos)

  2. Fictional travel platform

    Engineering

    2021.07 ~ 2023.02 (1 yr 8 mos)

Education

  • Fictional University

    Computer Science / 2015.03 ~ 2021.02

Start with your job post