Duke University, NC, USA

2023 - 2025

• Master of Engineering Management
• Finance | Marketing | Product Management | Student Consultant @ ALBERT

Indian Institute of Technology, Madras

2018 - 2023

• Bachelors in Mechanical Engineering + Masters in Robotics
• Minors in Computer Science + Minors in Artificial Intelligence

Microsoft Research, Redmond

2021 - Ongoing [Vibhav Vineet]

• Introduced a new Deep Learning training paradigm termed deepSIFT to tackle the vulnerability of traditional pipelines to distributional shifts.
• Proposed a differentiable SIFT module, a robust local representation, as a replacement for RGB-image inputs for CNN and ViT pipelines.
• Proposed approach surpassed state-of-the-art domain-generalization benchmarks by 10 points without loosing image properties.
• Paper submitted at CVPR, 2023.

Kotak Mahindra Bank

2022

• Developed a personalized banking-product recommendation framework based on user's web-search patterns and social profile.
• Introduced a ML framework, the first in Indian retail-banking industry, to identify the optimal marketing-channel and perform campaigns.
• Proposed framework improved revenues by 3% in simulations and is expected to be deployed in practise from 2023.

IBM Research

2020

• Developed a model-agnostic model-explainer for graph-neural-networks.
• Framework identifies porential nodes in a knowledge-graph responsible for a model's prediction.
• Method involves the identification of a mask for each graph-edge that aggregates information towards a prediction.
• Extended abstract submitted at ICM'21.

Interpretable Explanations & Quality Estimation for Endoscopic Videos

IIT-Madras [Prof. Chandrasekhar Lakshminarayan]

• Developed a two-stage heirarchial transformer to identify the correctness of a model-endoscopic procedure.
• Designed a framework that uses trained self-attention weights to identify key-frames of video procudures that are faulty.
• Working on developing a quality measure to evaluate a video-procedure of a surgeon with that of an expert.

Physics Informed neural Networks for Complex Engineering Simulations

AIDesign Pvt. Ltd. - IIT-Madras

• Foremost work on applying neural networks on physics based computational engineering with improved speed and accuracy.
• Improved computational speed by million times while achieving close to identical results as inefficient traditional methods.
• Attracted attention of media, investors & startups and led the tram to grow into a startup woth $4 million

Efficient Document Summarization using hybrid CNN-Transformers

Kellog School of Business, Northwestern University [Prof. Chaitanya Bandi]

• Developed a novel entity-extraction model to extract keywords from long-documents such as publications/articles.
• Sentence embeddings created using the proposed convolutional modules removes the quadratic-nature of self-attention mechanism.
• Provided insights on growth rate and hotness of research fields for the National Science Foundation, US Govt.