Experiences

Software Development Engineer-2

June 2019 - Aug 2021

Responsible for development of Adobe Scan & its image processing library MagicClean serving ~22.7 million monthly active users with over 50 million downloads. Integrated ML model and exposed API for document classification and text region detection.
Hackweek 2020: In a team of 5, implemented ML models for Adobe Scan to perform dewarping to remove scan curvature, handwriting stylisation, removal of scan artefacts like thumbs. Also designed an interpretable ML model for aspect based sentiment analysis of customer comments on Acrobat using LDA for topic modelling & BERT for sentiment analysis. Won honourable mention for 2 hacks.
Collaborated with the Magento team to work on data to text natural language generation using seller provided product details. Generated descriptions using the T5 transformer while controlling the linguistic style aspects based on buyer preference.

NLP Researcher

April 2020 - April 2021

Researched on fusion of audio-text contextual embeddings from CEO's speeches and used cross-modal attention fusion of Financial BERT text embeddings and audio contextual features to predict average and single day volatility in a multi-task setup.
Analysed the importance of leveraging the stock interrelations using graph nodes for stocks and earning calls and edges for wiki-data stock relations. Used semi-supervised Graph Convolutional Networks to extract the relational embeddings.
Empirically analysed gender bias in volatility prediction from mutli-modal data in CEO's speeches during quarterly Earnings Calls.

Mentor

October 2020 - Aug 2021

Mentoring a girl belonging to a low-income background from a small town of India on topics like vision setting, planning, decision making & job interviews

Undergraduate Research Assistant

Aug 2018 - March 2019

Researched on bio-medical image processing and skin cancer lesion images to propose an algorithm to detect & classify skin cancer at an early stage using Attention-Guided Deep CNNs.
Researched power quality disturbances to ensure reliable quality of power and classified them in voltage signals. Implemented an algorithm to classify disturbance events like voltage sag, swell, transients, harmonics, etc. using features extracted from CNN and Support Vector Classifier

Deep Learning Intern

June 2018 - Aug 2018

Implemented an image searching algorithm to find the most visually similar clothing items from a catalogue for an e-commerce client.
Researched object detection techniques including MaskRCNN, Fast and FasterRCNN. Used FasterRCNN in Tensorflow to detect and crop the item of interest from the query image.
Used Image Deep Ranking to train image similarity models on different categories of clothing items to understand the fine similarities between items using Keras.

Teaching Assistant

June 2018 - Sept 2018

Supported students in interview preparation on data structures & algorithms in C++, operating systems and DBMS.
Monitored progress and provided regular feedback to a group of 20 students including regular online support for doubts.

Undergraduate Research Assistant

Dec 2017 - May 2018

Proposed a model for surveillance using person re-identification to predict if two given videos are of the same person.
Trained a Siamese architecture with LSTM’s, CapsuleNets and dynamic routing based on agreement in Pytorch and plotted the results as cumulative matching curves.

Research Intern

May 2017 - Aug 2017

Proposed a machine learning based model to forecast the next day closing price of the Dow Jones Index.
Analysed various models including LSTM’s, 1D convolutions and dilated convolutions against the traditional models for financial time series forecasting.

Software Development Intern

June 2017 - July 2017

Designed and developed a web app with multi purpose business processing forms interlinked at various work steps that enables business stakeholders to make digital transformations in their operations.