Shreeram G. Subramanya
Shreeram G. Subramanya

Shreeram G. Subramanya

he/him

Cross-Modal World Models • Computer Vision • MS Data Science @ UB • Seeking PhD

I am a software engineer at White Hack Labs in New York, building AI-driven security tools for HackerGPT—a platform serving 10,000+ users. I completed my Master’s in Data Science from the University at Buffalo and worked as a Research Assistant at the Center for Unified Biometrics and Sensors (CUBS) Lab under Prof. Srirangaraj Setlur.

I’m actively seeking PhD opportunities in Computer Vision, focusing on cross-modal world models that learn physical properties through prediction discrepancy—the insight that when AI predicts cross-modal behavior and observes mismatch, the error reveals physical properties neither modality encodes alone.

News

  • Currently Seeking PhD opportunities in Computer Vision / Cross-Modal Learning for Fall 2026 intake.
  • May 2024 Presented and published "GestSpoof" at IEEE FG 2024 in Istanbul, Turkey
  • Apr 2024 Released GestSpoof Dataset for fingerprint presentation attack detection research
  • Nov 2023 Joined White Hack Labs as Software Engineer, working on HackerGPT
  • Feb 2023 – Nov 2023 Completed Research Assistantship at CUBS Lab, University at Buffalo
  • Feb 2023 Graduated with MS in Data Science from University at Buffalo

Research Interests

Cross-Modal World Models Audio-Visual Learning Video Understanding Biometric Security Intuitive Physics Authenticity Verification
**GestSpoof: Gesture-Based Spatio-Temporal Fingerprint Spoof Detection** IEEE FG 2024 • Istanbul, Turkey A novel approach using cross-modal prediction—learning that prediction error between expected and observed deformation reveals material properties. Achieved **83.84% F1 score** (10% improvement over static baselines).
Paper Dataset Details

Research Vision

I propose developing cross-modal world models that learn physical properties through prediction discrepancy. The core insight from my GestSpoof work: when a system predicts the deformation that should accompany applied pressure and observes a mismatch, that error reveals physical properties—elasticity, hardness, material authenticity—that neither modality encodes alone.

Aim 1: Bidirectional Audio-Visual World Models
Extending motion→sound and sound→motion prediction for material property inference.
Aim 2: Temporal Cross-Modal Prediction
Learning action dynamics through cross-modal consistency over time.
Aim 3: Authenticity Verification
Using prediction error as anomaly detector for AI-generated content.

Education

M.S. in Data Science, 2023
University at Buffalo, SUNY
B.E. in Computer Science, 2020
The National Institute of Engineering, Mysore

Experience

November 2023 – Present
Software Engineer @ White Hack Labs
• Working on HackerGPT, an AI-driven security platform serving over 10,000 users
• Designed Breach Outreach CRM: platform designed to bridge the gap between data breach discovery and proactive cybersecurity outreach. It identifies compromised organizations and streamlines communication with key stakeholders to facilitate rapid incident response and penetration testing.
February 2023 – November 2023
Research Assistant @ CUBS Lab, University at Buffalo
Developed GestSpoof framework for fingerprint presentation attack detection
March 2020 – July 2021
Data Consultant @ Cognitron Technologies
Worked on developing Educational CRM platform to boost student enrollment in public schools.

Highlights

GestSpoof HackerGPT

Interested in discussing research opportunities?

Schedule a Meeting Email Me