Google DeepMind GenAI - Project Astra Modelling Team.Project Astra explores the future of AI assistants that can process multimodal information, understand the context you’re in, and respond naturally in conversation.To know more about Google Astra, Please visit: https://deepmind.google/technologies/gemini/project-astra/

As part of Google’s Speech Research Team, I contributed to building large-scale Automatic Speech Recognition (ASR) models, with a focus on domain adaptation, data minimization through unsupervised learning, parameter-efficient fine-tuning, speech personalization, contextualization, and bias mitigation."Selected Work : 1. Reducing the WER disparity between African American English (AAE) Speech vs Standard American English (SAE) for Large ASR Models. Publication: Improving Speech Recognition for African American English with Audio Classification (https://ieeexplore.ieee.org/abstract/document/10447116/)2. Large Scale ASR Domain Adaptation using Self and semi-supervised learningPublications: 1. Large-scale asr domain adaptation using self-and semi-supervised learning (https://ieeexplore.ieee.org/abstract/document/9746719) 2. A Comparison of Supervised and Unsupervised Pre-Training of End-to-End Models (https://www.isca-archive.org/interspeech_2021/misra21_interspeech.pdf)3. Incremental layer-wise self-supervised learning for efficient speech domain adaptation on device (https://arxiv.org/abs/2110.00155)3. Improving Speech Personalization by Combining Personalized Language Models and ASR Models.Publications: Userlibri: A dataset for asr personalization using only text (https://arxiv.org/abs/2207.00706)Areas: GenAI, AI, LLM, Speech, ASR, Deep Learning, Un/Semi/Self-supervised/Active Learning, Federated Learning, Tensorflow.

Making GMaps awesome(r) !! Improved the Neural Classifier Model for inferring home/work location of a user using their location data by using clustering algorithm DBScan for detecting the home/work clusters of a user.

Understanding Harry Porter in an Alternate Universe! I worked with Prof. Carolyn Rose on improving the Natural Language Processing tasks on the fan-fictions written by mostly teen authors. The project is in collaboration with Yahoo and Tumblr and the final aim is to create chat-bots having personalities of the fictitious characters introduced in the fan-fictions. We focussed majorly on fan-fictions of Harry Porter, Friends and My Hero Academia.I primarily worked on improving the Coreference Resolution of the fan-fictions by combining features from Standford CoreNLP Statistical model and Neural Coref Models by AllenAI. Major challenges involved handling genders, aliases and inconsistent and informal writing techniques of authors.

Worked as the lead developer and designer for China Download Manager, a entirely new version of Adobe Download Manager with dynamic features for providing third party offers. The installer is successfully released in China and is expected to increase the revenue of the team multi-fold. This new dynamic installer can be configured to work with any number of third party offers and in many different modes. Responsible for the design, implementation and code/design review of the installer.Also, work involved working with multiple teams working in different team zones.

• Working in Core Product and Technologies Team of Adobe, India.• Independent developer for Adobe Download Manager software which manages downloads and distribution of Adobe Flash Player, Reader etc on all platforms and providing third party offers like google chrome etc resulting in around 50 million dollars of revenue every year.• Also worked on a side-project which involved extracting spatial information and exact geometry of objects using Microsoft Hololens. Segmented and reconstructed the mesh obtained for the concerned object, using the spatial mesh points information obtained through Hololens, and PCL (point cloud library), Future work involved refining that mesh using to extract the exact geometry. • Technologies used: C++, UWP, Unity, PCL, Visual Studio, Linux, OS X, Windows

• Working as an Intern in Product Engineering Team, [24]7 Innovation Labs.• Technologies used: Groovy, JAVA, MongoDB, Hazelcast, Elastic Search, HTML, CSS, Android.• Developed features for a web tool (server & client side) for analyzing huge amounts of data in real time and automate the actions for alerts.• Implemented rules framework in JAVA using MVEL as the expression language in order to make the system configurable for future.• Improved the performance to around 20% of the previous by implementing a Hazelcast cache layer.

• Technology Used : Google Glass, JAVA, Android Programming, Eclipse• Reviewed various Machine Learning Algorithms for Age Invariant Face Recognition.• Developed and tested innovative Android Applications based on QR Code Generation and Retrieval that have applications in the Medical Industry.• Professor : Asst. Prof. Dr. Shi Cao

Representative of BITS CodeChef Community, organized Inter-college Programming Competitions.

Lead a team consisting of 50 students to organize “Coding” events during APOGEE, Technical Fest of B.I.T.S. Pilani.Responsible for publicizing the events organized by A.C.M. by innovative ideasDesigned various Competitive Programming Questions and conducted events like Code Relay during APOGEE

Conducted a detailed study on Computer Networking in AMOL Pharmaceuticals Pvt. Ltd, Jaipur.Worked with the technical team and studied on various computer networking methodologies used.Identified and analyzed the problems faced by the company with the on-going network.Formulated and suggested methods and ways to minimize the cost, improve the speed and efficiency of networking in the company.