Interview integrity · on-prem

Every viva verified.

VivaGuard confirms the enrolled candidate is the one answering, and flags read-aloud or AI-generated answers. It runs entirely on your servers. No candidate audio ever leaves.

Book a demo See how it works
on-prem · zero data egress · your servers only
In their words

“We could never really tell who was answering. Was it AI, a friend standing in, or the candidate reading out AI-generated answers? VivaGuard confirms the enrolled candidate is the one speaking and flags answers that are being read rather than genuinely spoken. And it all runs on our own infrastructure, so no candidate audio ever leaves our environment. We wired it in with a simple API in a few days, and it indeed has strengthened our trust layer.”

Kumar Veetrag, CTO of enParadigm
Kumar Veetrag
CTO, enParadigm
enParadigm runs experiential, gamified learning & assessment programs that build capability for enterprises.
The problem

Remote voice interviews are easy to cheat.

Webcams and honor codes don't catch fraud that happens through the microphone, by someone who isn't the candidate, or who isn't really answering.

Impersonation

A stand-in sits the interview in place of the real candidate.

Reading AI answers

The candidate reads a pre-written or ChatGPT answer aloud instead of thinking.

Off-mic coaching

A helper off-camera feeds lines the candidate simply repeats.

Proctoring blind spots

Video watches the screen and room, not who is speaking, or how.

How it works

Two questions, asked of every answer.

Is this really the enrolled person, and are they genuinely answering? Three steps, in order.

01

Enroll

The candidate reads a random sentence in a short, timed window, a liveness step that captures a genuine, live reference voice.

02

Verify

Any later recording is matched against the enrolled voiceprint. Same person or not, with a clear accept/reject and a similarity score.

03

Exam mode

Scores speech naturalness (disfluencies, pauses, pacing) to tell genuine thinking-aloud from a read or fed answer.

Why VivaGuard

Built for institutions that can't send data away.

IDENTITY + NATURALNESS

Two signals, every cheat

Voiceprint matching plus speech-naturalness analysis, catching stand-ins, read-aloud, and coached answers.

PRIVATE BY DESIGN

Nothing leaves your server

No audio, transcript, or result ever leaves. Zero third-party calls, zero per-call cost.

SELF-HOSTED

One-command install

Docker, CPU-only, fully air-gapped after setup.

EASY TO INTEGRATE

A clean REST API

Three endpoints and a built-in console. Drop it into any interview or exam workflow.

Ready when you are

Bring integrity to every interview.

See VivaGuard run on your own infrastructure. No candidate data leaves your servers.

Book a demo

Frequently asked questions

What does VivaGuard do?

VivaGuard checks two things about every spoken answer in a remote exam or interview. First, whether the person speaking is the candidate who enrolled, using speaker verification against a live reference sample. Second, whether the answer is genuinely spoken rather than read aloud, using speech-naturalness analysis of disfluencies, pauses, and pacing.

What kinds of cheating does VivaGuard catch?

Impersonation, where a stand-in sits the interview in place of the real candidate. Reading AI-generated or pre-written answers aloud. Off-mic coaching, where a helper off camera feeds lines the candidate repeats. Conventional video proctoring watches the screen and the room, not who is speaking or how.

Does candidate audio leave our servers?

No. VivaGuard runs entirely on your own infrastructure. No audio, transcript, or result leaves your environment, which makes it suitable for institutions with data residency obligations under regimes such as the EU GDPR and India's DPDP Act.

How is VivaGuard deployed?

VivaGuard ships as a Docker deployment that runs on CPU only and is fully air-gapped after setup. It exposes a REST API with three endpoints and a built-in console, so it can be wired into an existing interview or examination workflow rather than replacing it.

How is speaker verification different from video proctoring?

Video proctoring watches the room and the screen through a webcam, so it detects a second face, a glance away, or a change of window. It does not analyse who is speaking or how the answer is delivered. Speaker verification answers the identity question directly, and speech-naturalness analysis answers whether the candidate authored the answer.

VivaGuard one-pager Deployment options isAI API documentation