Get started

For startups and scaleups hiring engineers

Your take-home test is grading ChatGPT

Gartner found 39% of candidates already use AI in the application process, and of those, 29% use it on assessment answers. There are public step-by-step guides for cheating on the major coding platforms. Savvanta screens the way cheating cannot survive: real tasks in a monitored sandbox, then a live conversation about the candidate's own code, graded with quoted evidence and confirmed by a human.

The numbers behind the cheating wave

39%
of candidates already use AI in the application process
Gartner survey, n=3,290, 4Q 2024
1 in 4
candidate profiles projected to be fake by 2028
Gartner, 2025
62%
of job seekers say they would ghost an employer, up from 37% in 2019
Indeed Ghosting in Hiring report, n=9,033, 2023

Peer-reviewed research now confirms the mechanism: candidates assisted by ChatGPT outscore others in asynchronous video interviews (Canagasuriam et al. 2025, International Journal of Selection and Assessment). Google, McKinsey, Cisco and Deloitte responded by bringing back on-site interview rounds. That works, and it costs your senior engineers a day per candidate. The alternative is an assessment where AI assistance does not help: explaining your own decisions, live, about code you just wrote.

What a Savvanta technical screen looks like

  1. Role brief in, assessment out. Sandboxed coding tasks generated for your stack, approved by you before any candidate sees them.
  2. Proctored attempt with explicit consent, an equipment check, one take.
  3. Live technical conversation. An AI interviewer asks the candidate to walk through their own solution, probe trade-offs, and handle a change request in real time.
  4. Evidence-graded. Fixed criteria scored with quoted moments and playable clips.
  5. A human confirms every decision. Integrity flags never auto-reject.

What screening actually costs

What a screened candidate costs today (published pricing and estimates)

Automated test platformsAutomated test platforms: ~$20-25~$20-25Your senior engineer, 90-min screenYour senior engineer, 90-min screen: ~$150-250 (modeled)~$150-250 (modeled)Human interview-as-a-serviceHuman interview-as-a-service: ~$200-450 (estimated)~$200-450 (estimated)
Published per-attempt pricing: HackerRank $20, CodeSignal $20, CoderPad $25 (vendor pricing pages via interviewcost.com). Engineer cost is a stated model. Interview-as-a-service figures are third-party estimates; no public price list exists. Savvanta prices between the automated and human tiers, with the live conversation included.

Why evidence beats resumes

What actually predicts job performance (meta-analytic validity, r)

Structured interviewsStructured interviews: r = .42r = .42Work samplesWork samples: r = .33r = .33Unstructured interviewsUnstructured interviews: r = .19r = .19
Source: Sackett, Zhang, Berry and Lievens 2022, Journal of Applied Psychology, the largest modern meta-analysis of selection methods. Blue marks the methods Savvanta assessments are built on. Estimates carry wide credibility intervals and vary by role.

The strongest predictors of job performance are structured work samples and structured interviews. A CV screen plus an unstructured chat is the weakest common combination. This is not our claim; it is the largest meta-analysis in personnel psychology.

See a live assessment end to end. Twenty minutes, real tasks, a real evidence report.

See a live assessment

Frequently asked questions

How is this different from HackerRank or CodeSignal?

Those are test libraries you administer. Savvanta generates role-specific tasks for your stack, runs a live conversation about the candidate's own code, and puts a named human review on every decision.

Can candidates still use AI?

On the written portion the sandbox is monitored and paste events are recorded as observations, never as automatic penalties. The live conversation is the real filter: explaining code you did not write, under follow-up questions, does not survive contact.

What does the candidate experience look like?

Consent first, an equipment check with a proven upload, one take, per-task receipts, and a status page with a dated decision. No silent rejections.

Is it GDPR compliant?

EU-hosted in Germany, explicit consent before any capture, retention limits aligned to legal windows, and a human review on every decision consistent with Article 22. See our GDPR guide for the full picture.

What does it cost?

Per candidate screened or per successful hire. No seats, no annual lock-in.