Ecosystem for creating & trading in-game assets using Generative ML

A secondary marketplace to create and buy/trade/sell video game assets securely.

It employs generative ML diffusion models to enable creation and content moderation of 3-D assets that can be deployed in game


VIrtual STereo based Augmentation for depth estimation in automated driving (NeurIPS 2022 Autonomous Driving)

Partners: Bin Cheng, Kshitiz Bansal, Professor Dinesh Bharadia & Gaurav Bansal (UC San Diego & Blue River Technology)

Improves the State-of-the-Art monocular depth estimation by 8% in automated driving using a novel Virtual Stereo camera approach.

Based on a custom simulated dataset created on CARLA mimicking the limited real life computer vision KITTI dataset.


Online Platform for locally-sourced unique artisanal products

An online marketplace for local/home-based businesses to sell unique customizable home-made items.

Built as a React web app with a MongoDB backend

Realistic Colorization of Portrait Line Art

10-745 Scalability in Machine Learning, Carnegie Mellon University

Partner: Julia Shieuh, Georgia Tech

Explores line-art colorization of human faces using Generative Adversarial Networks. Based on a pix2pix model on top of a custom dataset constructed using different lineart generators

GAN-based Unsupervised Clickbait Style Transfer

10-617 Intermediate Deep Learning, Carnegie Mellon University

Partner: Sayani Kundu, Carnegie Mellon University

Used CycleGAN based models, in particular StyleTransformer, to convert between factual news titles (non-clickbait) and viral hyperbolic and sensational headlines (clickbait)

We incorporate semantic and syntactic features of clickbaity text into the existing model to aid in transforming the style of a text while preserving legibility and meaning.

Resource Sharing in Networks using a blockchain-based Cap & Trade scheme

Heinz College of Information Sciences, Carnegie Mellon University

Advisors: Ramayya Krishnan (Dean of Heinz College), Wang Hai, Carnegie Mellon University

It envisions "Cap and Trade" using a distributed ledger as a universal strategy to share resources in different domains among users fairly and optimally.

Decentralizing Aadhaar, India's Identity Platform

Independent Research on introducing a blockchain layer to Aadhaar Authentication

Aadhaar provides a unique biometric identity to 1.2 billion residents of India.

This proposal aims to increase the integrity and privacy of Aadhaar by providing an immutable decentralized layer on top of the current architecture.

Natural Language Processing in the Indian Context

NITI Aayog, the Indian govt. thinktank & International Innovation Corps, University of Chicago

Research on the development of Natural Language Processing (NLP) applications in 18 Indian Vernacular languages

The internal paper is being utilized to construct an Access Point Repository for Indian NLP applications for various use cases.

GoalKick — A Goal Scheduling Application employing intelligence automation

The Indus Entrepreneurs (TiE) The Young Entrepreneur Program

Application built and pitched as part of a program conducted at Indian Institute of Technology, Delhi

The app helps in maintaining personal and professional targets and finding time for new and pre-existing goals customized to the user's schedule.