Research and Patents
Our patents
Protecting innovation, powering trust
Our patents protect the core technologies behind our products, from media analysis and audio forensics to content generation and advanced AI systems. Each patent reflects our commitment to creating original, practical solutions that strengthen trust and integrity in digital systems, giving investigators and enterprises confidence in solutions that are future-ready, court-ready, and trusted worldwide.
Patent | Application Id. | Filed On | Published On | Inventors |
|---|---|---|---|---|
A system and a method for media analysis |
202421080389 |
22-10-2024 |
03-01-2025 |
Raghu Sesha Iyengar, Abhijeet Zilpelwar, Ankush Tiwari, Asawari Bhagat, Srivallabh Mangrulkar |
A method and system for identifying a speaker of interest in an audio |
202421080345 |
22-10-2024 |
03-01-2025 |
Raghu Sesha Iyengar, Abhijeet Zilpelwar, Ankush Tiwari, Bijay Singh, Naveen Sharma |
A method and system for content analysis |
202421080425 |
22-10-2024 |
03-01-2025 |
Raghu Sesha Iyengar, Abhijeet Zilpelwar, Ankush Tiwari, Mallampalli Kapardi, Harshal Bhore |
A system and a method for audio analysis |
202421080381 |
22-10-2024 |
22-11-2024 |
Raghu Sesha Iyengar, Abhijeet Zilpelwar, Ankush Tiwari, Prabakaran Nandakumar, Anant Dhok |
A method and system for content generation |
202521013377 |
17-02-2025 |
28-02-2025 |
Raghu Sesha Iyengar, Abhijeet Zilpelwar, Ankush Tiwari, |
Published papers
Grounded in research, proven in benchmarks
pi-labs is committed to advancing the science of digital forensics and investigative AI. Our research is published in leading international conferences and journals, covering breakthroughs in deepfake detection, generative source attribution, video KYC security, mobile-enabled AI, and novel face embeddings. These contributions push the boundaries of AI while reinforcing transparency, rigor, and global relevance. this ensures our solutions consistently align with the highest standards of accuracy and reliability
Paper | Description | Authors | Paper DOI |
|---|---|---|---|
Forensic Challenges in Face Manipulated Videos |
This paper presents a technique for detecting deepfakes in both image and videos, including face-swaps and lip-sync manipulations, achieving state-of-the-art accuracy on standard benchmark datasets. |
Naveen Sharma, Kapardi Mallampall, Raghu Sesha Iyengar | |
Fingerprinting Generative Audio Sources |
This paper presents a technique for detecting audio deepfakes by uncovering unique generative fingerprints corresponding to different synthesis models. The proposed method achieves state-of-the-art performance on standard benchmark datasets. |
Vardhini P, Naveen Sharma, Harshal Bhore, Raghu Sesha Iyengar |
Accepted in 2025 8th Artificial Intelligence and Cloud Computing Conference. |
Enhancing Trust in VideoKYC: Deepfake Detection and Source Attribution |
This paper refines deepfake detection models with a specific focus on videos KYC applications. |
Srivallabh Mangrulkar, Aryan Karande, Abhijeet Zilpelwar, Raghu Sesha Iyengar |
Accepted in 16th EAI International Conference on Digital Forensics & Cyber Crime |
Deepfake Detection as a Service: Enabling Trust on Mobile Devices |
This paper presents a mobile application for deepfake detection, which leverages a backend detection engine exposed via APIs to deliver efficient on-device analysis. |
Abhijeet Zilpelwar, Ansh Tiwari, Raghu Sesha Iyengar |
Accepted in IEEE 5th Cyber Awareness and Research Symposium 2025 (CARS'25) |
Learnable Rotary Position Embeddings for Face |
This paper introduces a novel technique for deriving face embeddings, enabling improved face identification and matching, and achieving state-of-the-art accuracy. |
Vardhini P, Raghu Sesha Iyengar | |
Graph Neural Networks for Video Device Identification |
This paper employs graph-based techniques on video metadata to identify the device on which a video was originally captured. |
Raghu Sesha Iyengar, Vaibhav Kumar |
Accepted in 16th EAI International Conference on Digital Forensics & Cyber Crime |
A framework to ease dApp development and adoption |
This paper introduces a platform designed to simplify the development of distributed applications on blockchain. |
Naveen Sharma, Bijay Singh, Abhijeet Zilpelwar, Ankush Tiwari, Raghu Sesha Iyengar |
Accuracy Metrics Comparison – Video Model
Different Deepfake Detection Models - Video | Public dataset - DFDC (AUC – Area under curve) | Public Dataset Face Forensics++ (Accuracy) |
|---|---|---|
ViT with Distillation |
0.978 |
- |
Selim EfficientNet |
0.972 |
- |
Convolutional ViT |
0.843 |
0.93 |
Conv. Cross ViT Eff.Ne |
0.947 |
- |
Efficient ViT |
0.919 |
0.83 |
Conv. Cross ViT Wodajo CNN |
0.925 |
0.83 |
Authentify Video Model |
0.998 |
0.97 |
*Lower the value better the accuracy
Accuracy Metrics Comparison – Audio Model
Different Deepfake Detection Models - Audio | Public dataset – ASVSpoof2019(EER) |
|---|---|
RawGAT ST |
1.06 |
ResTSSDNet |
1.64 |
MCG-Res2Net50+CE |
1.78 |
ResNet18-LMCL-FM |
1.81 |
LCNN-LSTM-sum |
1.92 |
Capsule network |
1.97 |
Resnet18-OC-softmax |
2.19 |
Authentify Audio Model |
0.54 |
*Lower the value better the accuracy