Leveraged LLM (Large Language Models) like GPT 4V, and GeminiAI, to develop cutting-edge multimodal models tailored for fashion data, Compared the results by finetuning(using LoRA) open-source models like LLama2 and Idefics to create an aesthetic classifier which performed better than vision transformer techniques by 15% point.Led the creation of a groundbreaking Visual Search Product, employing Transformer based computer vision techniques and integrating contrastive loss using tfSim and CLIP, to enhance the discoverability of garments for customers, elevating the overall user experience and increasing engagement by 10%.

• Led the designing and deployment of the in-house Recommendation Decision System utilized by the company to suggest stickers and GIFs on various messaging platforms, increasing user engagement by 22%.• Utilized Tensorflow lite to train and ship Deep Neural Network-based multi-class classifiers on-device as part of the General Recommender, which aims at suggesting appropriate content based on situations•Create a unique On-device Storage System to store user interaction with the help of Bloom Filters to train models as part of Privacy-Preserving Machine Learning. • Deployed a Deep Reinforcement Learning model on-device to rank stickers based on the user's recent history, increasing personalization of content.• Led a team of Data Science interns to create a pre-campaign predictive time series analysis platform, used by Holler's Marketing team with branded campaigns and improving campaign performance by 300%.• Designed a Distributed AI approach based on Dask to train deep learning models and experiment with Incremental learning.• Led the entire company to launch an in-house website to display machine learning results from classifiers for themes, sentiment, emotions, and intent and devised a Mechanical Turk experiment achieving an accuracy of 90%.

• Created a Machine Learning architecture to investigate and analyze company data on Databricks using PySpark• Designed and deployed novel Learn-to-Rank algorithms based on Reinforcement Learning methods, increasing the share rate by 3%• Implemented novel Interleaving algorithms using on-device capabilities to boost the visibility of new stickers.• Implemented rigorous A/B testing to monitor various KPIs and put in place optimization methodologies and Testing playgrounds to help test future models and compare.

Working on various projects based on Deep learning, Computer vision, and Game theory.• Pioneered research and implemented various Generative Adversarial Networks (GAN).• Created a GPU-compatible code for PassGAN migrating from Pytorch to Tensorflow.• Designed GAN applications in the fields of music and medicine, which lead to the creation of MIDIGAN and GeneGAN respectively.• Developed an auto-generated visual art that can be played on the piano creating a novel GAN application, MIDIGAN.• Achieved 2-fold improvement for music generation over the existing method, MuseGAN.• Modified Wasserstein Generative Adversarial Network(WGAN) create Expressed sequence tags from virus strands extracted from the NCBI virus database; a first text-based application of the system on genetic data, GeneGAN

• Assistant for the Artificial Intelligence course, leading discussions in statistics and machine learning concepts like clustering, regression, classification, and Natural Language Processing.

• Helped develop domocracyos.in, a discussion platform for democratic decision-making using HTML/CSS.• Hosted the site using Docker and Nginx and used Google Analytics to analyze user traffic.• Maintaining the site and using SEO techniques and google analytics to create a user base.

• Lead a team of developers to design an algorithm to tackle Replay Attacks on wireless sensor networks.• Implemented Bloom filter mapping reducing memory consumption to 10%.• Solved the challenge of deploying the algorithm at the Link Layer for the FlexiSec architecture.

• Cofounded the startup and designed the Crossbar app and engineered the navigation along with its UX/UI.• Crossbar app aimed at creating a general platform for the booking of Astro turf spaces throughout India.