Veritas Engine v3 -- Live

The End of
Remote Fraud.Continuous identity verification that runs entirely in your browser.

A remote senior dev costs $30,000 in risk if they fake the interview with AI. Deep-Check validates identity continuously -- not just at login. Six independent detection layers. Zero biometric data leaves the device.

No credit card. No signup. Runs in your browser.

rPPG Blood Flow
FACS Micro-Expressions
EfficientNet-B4 Pixel
CNN v2 Blendshape
Keystroke Biometrics
Session Continuity
VERITAS ENGINE ACTIVE
ISO 30107-3 CompliantAUC 0.9999EER 0.10%GDPR NativeEU AI Act ReadyClient-Side InferenceZero Biometric Storage6-Layer Detection Stack195 Countries SupportedBrowser-Based ONNXPrivacy by DesignContinuous VerificationISO 30107-3 CompliantAUC 0.9999EER 0.10%GDPR NativeEU AI Act ReadyClient-Side InferenceZero Biometric Storage6-Layer Detection Stack195 Countries SupportedBrowser-Based ONNXPrivacy by DesignContinuous Verification
0.10%
Equal Error Rate
ISO 30107-3
0.9999
AUC Score
155K test images
6
Independent Detection Layers
Bayesian fusion
0 B
Biometric Data Sent to Servers
Client-side only
How It Works

Four layers between your
organization and fraud

Deep-Check does not verify once and forget. It watches the entire session, fusing six independent signals in real-time.

Step 01

Face Enrollment

The candidate shows their face. MediaPipe extracts 478 landmarks. No image is stored -- only a local embedding used for session matching.

Step 02

Liveness Verification

Six layers run simultaneously: rPPG heartbeat, FACS micro-expressions, EfficientNet-B4 pixel analysis, CNN blendshape classifier, blink physics, and lighting reflex detection.

Step 03

Behavioral Biometrics

Keystroke dynamics capture flight time, hold time, and typing rhythm. Code velocity analysis flags AI-generated copy-paste patterns in real-time.

Step 04

Continuous Session Score

All signals fuse through a Bayesian ensemble in logit-space. The trust score updates every second. Anomalies trigger instant alerts to the hiring manager.

The Detection Stack

Built to be unfakeable

Not one model. Not one check. Six independent layers that would each need to be defeated simultaneously.

rP

rPPG Heartbeat Coupling

Extracts blood volume pulse from facial video. Real skin shows cardiac rhythm; deepfakes and static images produce flat or synthetic signals. Weighted at 0.30 in the Veritas ensemble.

Bk

Keystroke DNA

Flight time, hold time, and rhythm patterns are as unique as a fingerprint. Impossible to replicate in real-time.

FA

FACS Micro-Expressions

Action Unit analysis detects involuntary facial movements that deepfake renderers cannot reproduce at the correct temporal frequency.

B4

EfficientNet-B4 Pixel Forensics

18.6M parameter model with multi-scale Laplacian frequency branch. Trained on 155K images across 7 datasets. AUC 0.9999, EER 0.31%. Anti-leak verified with hash deduplication and zero train/test overlap.

Cv

CNN v2 Blendshape

Separate classifier operating on blendshape coefficients extracted from the face mesh. Catches swap artifacts invisible to pixel-level analysis.

Cd

Code Forensics

Detects LLM-generated code by measuring perplexity scores, typing cadence, and paste-vs-compose velocity anomalies during live coding interviews.

Sc

Session Continuity

Continuous re-verification ensures the person who started the session is the same person throughout. Face-swap mid-session triggers instant alerts.

Competitive Analysis

Deep-Check vs Onfido

Onfido verifies who someone is once, at login. Deep-Check verifies who someone is continuously, during the entire session.

CapabilityDeep-CheckOnfido
Identity Check at LoginYesYes
Continuous Session MonitoringReal-timeOne-time only
Deepfake Detection6-layer stackBasic liveness
Heartbeat Signal AnalysisrPPG couplingNot available
Behavioral BiometricsKeystroke DNANot available
AI-Generated Code DetectionFor interviewsN/A
Document Forensics (ELA/EXIF)Built-inAdd-on
On-Premise DeploymentEnterpriseCloud only
No Biometric StorageClient-sideSends to servers
Starting PriceFree tier1,500+/month
GDPR + EU AI ActFullPartial
Product Module

Document Forensics

Detect manipulated images, deepfakes in documents, and falsified metadata in grant applications, dossiers, and document verification workflows.

ELA -- Error Level AnalysisEXIF Anomaly DetectionAI Image SignatureNoise Forensics
Analyze a document
EL

Error Level Analysis

Detects edited regions by comparing JPEG compression artifact levels across the image

EX

EXIF Metadata

Identifies editing software signatures (Photoshop, GIMP, Canva) embedded in file metadata

AI

AI Image Detection

Distinguishes real photographs from AI-generated images across Midjourney, DALL-E, and Stable Diffusion

Enterprise

Join the Waitlist

Get early access to Deep-Check for your organization. Continuous identity verification, on-premise deployment, and dedicated support.

No spam. We only email when your access is ready.