Deep learning for Chest Disease Classification

We developed a Deep Neural Network model for multi-label thorax disease classification on chest X-ray images (CheXpert Dataset). We experimented with DenseNet-121, ResNet-152 and Auto-ML architectures.

Traffic Sign Detection and Classification

We developed RetinaNet (Faster-RCNN) based modek for Traffic Sign (140 sign classes) detection and classification. Achieved recall of 94%

Pedestrian Crossing and Guardrails Detection

Developed Mask-RCNN model for Guardrails and Pedestrian Crossing Segmentation.

Unsupervised Cross-Domain Image Generation using GANs

Explored unsupervised domain transfer from an image of domain S to an image of domain T without any mapping data.We created models for four set of domains, analysing the adaptation of the architecture on different distributions and evaluated the models. We achieved similar results for digits datasets but not faces, finding that compute resources and training strategies play a vital role in successful implementation.

Leveraging Social Media and Threat Intelligence Data Feeds to Detect Early Signs of Internet Abuse.

Developed a ML system, based on graph streaming analytics, that continuously monitors social media feeds,and multiple threat intelligence data feeds to detect early-stage evidence of cyber attacks.

Unsupervised Word Sense Disambiguation.

Project aims to disambiguate the meaning of target word in a given para- graph or paragraphs. Algorithm achieved 80% accuracy on SENSEVAL-1 and SENSEVAL-2 datasets.

WalmartDec 2016 - July 2018

Software Engineer III, Machine Learning Platform

Designed and implemented micro-services to create on-demand cloud servers (IDEs) (Rstudio, Jupyter, Zeppelin etc).
Designed and implemented system to periodically execute long running machine learning jobs with on-demand compute resources.
Created Python and R API to manage package and library dependencies required for running ML models through Hadoop Streaming.
Created containerized Scoring Servers for various Data Science applications, reducing the maintenance overhead.

IBMMay 2015 - Dec 2016

Staff Software Engineer, IBM Data Science Experience

Designed and implemented Real-Time Scoring APIs for IBM Watson ML platform powered by Apache Spark.
Created an Apache Spark powered execution engine of an Internal Entity Resolution Language.

Yahoo!July 2014 - May 2015

Software Developer, Search Analytics and Reporting Platform

Created and maintained data pipeline for generating partner insights reports for Yahoo broadband partners.
Improved data processing time using Apache Hive, Apache Impala and Apache Oozie.

Yahoo!Feb 2014 - May 2014

Software Developer Intern, Yahoo Answers!

Simplified Hadoop based analysis of Yahoo answers grid dump data, and created data pipeline to integrate server logs in the same hadoop cluster.