Loading...
Loading...
A real-time security layer running directly on your device to verify whether the person you are speaking with on video calls is genuinely human.
This project is building a real-time security layer—like an antivirus for video calls—that runs directly on your device to verify whether the person you are looking at and speaking with is genuinely human.
Instead of simply guessing from pictures, it continuously checks real-world physical rules that AI cannot easily fake, such as whether the face naturally moves with the skull bone, how skin reflects light, and whether spoken words match exact lip movements. Its purpose is to run quietly in the background during video calls, online banking, job interviews, and dating apps, instantly alerting you if someone is using a face swap, cloned voice, or synthetic avatar to impersonate a person.
An end-to-end multi-stage neural pipeline from live biometric ingestion to 768-D latent projection, 3D manifold learning, and real-time 3-class probabilistic verification.
Biometric Feed
DINOv2 ViT
Latent Space
Biomechanical
Classification
Accurately separates authentic camera captures from fully synthetic AI creations and targeted face-swaps or deepfakes.
Maps high-dimensional representations into 3D coordinate space, revealing clear spatial separation between authentic and synthetic clusters.
Pairs every classification outcome with the closest matching images from training benchmarks to ground decisions in verifiable visual evidence.
Let's discuss how this can work for you.
Get in Touch