May 2020 – Aug 2020 · 3 mos · Sunnyvale, California, United States

• Accepted for Oral Presentation at AMLC Workshop 2020, the public version available soon (currently under submission at a top-tier conference)
• Leveraging data from multiple domains and tasks to efficiently boost the performance of low-resource NLP task
• Solving few-shot Dialog State Tracking problem using meta-learning algorithms on top of a BERT based model
• Extensive experimentation to highlight evidence in favor of meta-learning over multi-task learning in the low data setting

Sep 2019 – Jan 2021 · 1 yr 4 mos · Greater Pittsburgh Region

• Working on a robust pipeline for Handwriting Recognition on limited and noisy data to digitize maintenance log forms
• Using deep learning models for two-stage approach including word segmentation followed by word recognition

May 2019 – Sep 2019 · 4 mos · Greater Delhi Area

• Setup a Natural Language Processing(NLP) pipeline for the development of chatbot based product
• Used ensemble of RASA Library and rule-based system to solve entity extraction and intent classification problem
• Implemented a system for textual analysis of ASR transcripts of customer-sales associate interactions for sentiment analysis, aspects extraction, and dialogue modeling

May 2018 – Jul 2018 · 2 mos · Gurugram, Haryana, India

• Implemented Cache-like system for multi-client server application using Redux library; optimized pipeline by inte-grating faster incremental Trade Alarms updates.
• Added React Components to implement features like live feed, query status; detected risk limit errors in trade logs.
• Implemented robust python back-end server using Flask SocketIO for two way communication with dashboard; Trade PNL visualizations using Highstocks library with appropriate flags for errors.

May 2017 – Jul 2017 · 2 mos · Switzerland

• Persona: High Performance Bio-informatics framework (May, 2017 - July, 2017) : Research Intern involving making high speed software at Very Large Scale Computing Laboratory, EPFL, Lausanne, Switzerland under Prof. James Larus
• Designed an Op for computing gene coverage values, variant calls in Tensorflow based dataflow.
• Analysed the open souce software & integrated it in pipelined framework with AGD Datastructure.
• Achieved twice speedup over standard tools available for gene coverage value using efficiency.