cv

Senior Research Scientist (L5) | Team Lead, Efficient GenAI

Snap Research

Lead Efficient Generative AI Team, directing the technical strategy for Snap’s hardware-agnostic generative stack.

• Elastic Supernetworks: Spearheading the development of SnapGen++, a pioneering Elastic Diffusion Transformer (DiT) framework.
• Efficiency Frontiers: Architecting the next generation of SnapVideo models, achieving superior generation quality at significantly lower inference compute.
• Mobile Video Pioneers: Directing the team in the design of high-performance DiT models capable of real-time streaming video generation on-device.
• Strategic Mentorship: Leading a team of research scientists to bridge the gap between foundational research and production-scale AI deployment.

Research Scientist (L4)

Snap Research

Developed foundational generative architectures that transitioned from academic highlights to core Snapchat product features.

• SnapGen (Patent-Pending): Authored the landmark SnapGen paper (CVPR 2024 Highlight); orchestrated the architecture for which Snap Inc. filed a United States Patent.
• Large-Scale Product Impact: Engineered SnapVideo v2, powering numerous AI Lenses used by hundreds of millions of Snapchat users daily.
• Scholarly Leadership: Mentored over 10 Research Interns, resulting in prolific output at top-tier conferences (CVPR, NeurIPS).

Research Fellow

Microsoft Research

Advanced the field of Extreme Multi-label Classification (XMC) for industrial recommendation systems.

• Developed Parabel and SwiftXML, core algorithms that achieved a 13.6% increase in CTR for Bing Ads.

Software Engineer

Microsoft

Optimized the Interactive Service Hub for Dynamics CRM, reducing page load times by over 50%.

Ph.D.

Boston University

Electrical & Computer Engineering

• Dissertation: Novel Neural Architectures & Algorithms for Efficient Inference
• Committee: Profs. Venkatesh Saligrama (Advisor), Brian Kulis, Alexander Olshevsky, Kilian Q. Weinberger, and Prateek Jain.

M.S.

Boston University

Electrical & Computer Engineering

B.Tech.

Indian Institute of Technology

Computer Science

• Ranked 4th out of 80 students in the Computer Science Class of 2014.

ELITE: One Model, Many Budgets: Elastic Latent Interfaces for Diffusion Transformers

Snap Inc.

Patent Status: Approved for Filing (Sept 2025)

Sprint: Sparse-Dense Residual Fusion for Efficient Diffusion Transformers

Snap Inc.

Patent Status: Approved for Filing (Sept 2025)

DenseDPO: Fine-Grained Temporal Preference Optimization for Video Diffusion

Snap Inc.

Patent Status: Approved for Filing (May 2025)

SnapGen: Taming High-Resolution Text-to-Image Models for Mobile Devices

Snap Inc.

Patent Status: US Patent Filed (Nov 2024)

BitsFusion: 1.99 bits Weight Quantization of Diffusion Model

Snap Inc.

Patent Status: US Patent Filed (May 2024)

Sf-V: Single Forward Video Generation Model

Snap Inc.

Patent Status: US Patent Filed (May 2024)

Mind the Time: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis

Snap Inc.

Patent Status: US Patent Filed (Nov 2023)

AsCAN: Asymmetric Convolution-Attention Neural Networks

Snap Inc.

Patent Status: US Patent Filed (Nov 2023)

TextCraft: Your Text Encoder Can be Image Quality Controller

Snap Inc.

Patent Status: US Patent Filed (Nov 2023)

CVPR Highlight Award

CVPR

Research on SnapGen and Snap Video selected for Highlight presentation (Top 3-5% of global submissions).

NeurIPS Spotlight Award

NeurIPS

Recognized for high-impact contributions to Video Diffusion (DenseDPO).

CVPR Highlight Award

CVPR

ICML DyNN Workshop Spotlight

ICML

NeurIPS Spotlight Award

NeurIPS

Recognized for high-impact contributions to Selective Classification.

Top 10% Reviewer Recognition

NeurIPS

Rafik Hariri Graduate Student Fellowship

Boston University

Dean's Ph.D. Fellowship

Boston University

IIT-JEE All India Rank 1761

Secured a rank in the top 0.4% of 450,000+ candidates.