Hire engineers whothink ahead.

Ducker gives every candidate the same challenge and the same AI agent. Getting working code out of it is the easy part. We score the judgment that goes on top.

Get early access See how it works

backend-assessment / payment-service42:18 remaining

codebaseagent chatdiff 12terminaltests

payment.service.ts

38  async capturePayment(orderId: string) {
39    const order = await this.orders.findOne(orderId);
40
41    if (!order) throw new NotFoundException();
42
43  + await this.lock.acquire(orderId);
44  + try {
45      return await this.gateway.capture(order);
46  + } finally {
47  +   await this.lock.release(orderId);
48  + }
49  }

main*Ln 46, Col 19 UTF-8 TypeScript

REAL CODEBASESAI AGENTS ALLOWEDPROCESS-AWARE SIGNALSPRODUCTION JUDGMENT

Stack agnostic / real environments

Built for the stack your team already ships.

NestJS
Express
Fastify
Go
Python
React
Vue.js
Next.js
Nuxt
Flask
Django
Rust

Meet Qwack

Qwack is our fine tuned model, not a flagship. It's lazy, it isn't proactive, and it can make mistakes. What it writes usually runs, which is not the same as code you would want to own.

The candidates worth hiring get code that passes review anyway. They doubt it, they push, and they stand behind what ships. That is what we measure: not which agent they had, but what they got out of it.

Doubt
Did they read what came back, or accept it?
Direction
Did the prompts sharpen, or repeat?
Standards
Did they settle for working, or hold out for better?
Ownership
Would they defend this code in review?

Pick the track. We record who caught what.

Every task runs in a real repository with an AI agent beside the candidate. Ducker logs which concerns the candidate raised, which the agent raised, and which nobody did.

The endpoint works. The question is whether it holds.

Task · Reservation endpoint

Race conditionCandidateTransaction boundaryAgentConcurrent testCandidate

Hidden tests · 14 of 14 passed

Oversellingtwo bookings take the same seat unless someone sees it comingIdempotencya retried request must not charge the customer twiceSee the backend track

Evidence, not a score out of ten.

What lands on the candidate record when the session ends.

Candidate report

Every claim links back to the moment it happened.

Hidden tests

The edge cases the brief never mentioned, run on every submission.

Agentic code review

Structure, naming, error handling, the things a senior flags in a pull request.

Follow-up interview

Qwack asks why they made the calls they made, in their own code, while it is fresh.

Your own challenges

Bring a task from your repo and run it on your stack.

Ducker fits the pipeline you already run.

Send a challenge from your ATS and get the report back on the candidate record. Your recruiters never open a second tool.

API and webhooks

Plans & Pricing

Start with no subscription and pay only when a candidate finishes. Move to Growth once hiring becomes routine.

On Demand

For teams hiring a few engineers a year.

then $24 per completed assessment

Start free

Every task in the library
All supported runtimes and stacks
Full candidate report and session replay
One open role, single seat

Growth

For teams running assessments every week.

$149/mo

10 included, then $15 each

Choose Growth

Everything in On Demand
Unlimited open roles
Up to 15 team members
Author custom tasks with AI
AI follow-up interview
ATS integrations, API and webhooks

Enterprise

For hiring across many teams at once.

Talk to sales

Everything in Growth
No per-assessment fees
Unlimited team members
SSO, audit log and priority support

No credit card to start. On Demand charges only for assessments a candidate completes.

FAQ

The rubber duck talks back now.

Early access / now open

Meet the developers you'll want to build with.

Get early access

No credit card · Pay only for completed assessments

Hire engineers whothink ahead.

Built for the stack your team already ships.

Meet Qwack

Doubt

Direction

Standards

Ownership