Hi, I am Ashwini

I am a D|

Ashwini Marathe

M.S. (Data Science) student at Duke University

I am a Graduate student at Duke University studying Data Science. My passion drives data-driven changes, using machine learning techniques to uncover trends and suggest data-driven recommendations. Through my academic projects, I have had the opportunity to implement various predictive models like regression, classification using statistical and deep learning methods. With my strong problem-solving skills and ability to collaborate with diverse teams, I am ready to deep dive into challenging analytical problems.

Problem Solving
Leadership
Team Work
Communication
Hard Working
Fast Learner

Experiences

1
Data Science and Advanced Analytics Summer Associate
Capital Group

Jun 2020 - August 2020, Los Angeles, California

Capital Group is an American financial services company

Responsibilities:
  • Implemented a hybrid news recommender for financial advisors in Python based on their implicit interests and article features
  • Enhanced the model to alleviate cold start problem prevalent for high-frequency news articles platform
  • Suggested data-driven process improvements for the institutional client services team to optimize client engagement
Duke University

Sep 2019 - Present, Durham, North Carolina

Graduate Research Assistant

Jan 2021 - Present

  • Impute, analyze and do causal inference on survey data collected from PhD students graduating from Duke
  • Report analysis and statistical inference to the team and help with data-driven decision making
Graduate Research Assistant

Aug 2020 - Present

  • Analyzing Native American dislocation using census information and Legislative district reference maps in GIS and Python
Teaching Assistant

Aug 2020 - Oct 2020

  • Teaching Assistant for Data Infrastrucuture course under Fuqua Business School
  • Solved doubts and corrected homeworks and quizzes for a class of 160 students
Software Intern for Clinical Research Project

Sep 2019 - Dec 2020

  • Created and maintained user-friendly Computerized adaptive tests (CAT) on Concerto platform for pain measurement and management
  • Designed test automation scripts to enable a faster develop-test cycle
  • Created data dictionaries from the test configuration files to be used by statisticians and clinicians using Python scripts
2
3
Research Engineer
Conduent Labs

Jul 2017 - Jun 2019, Bengaluru, India

Responsibilities:
  • Enhanced an algorithm to predict temporal and spatial dependence of parking space occupancy using statistical methods, based on sparse data collected by enforcement officers and payment data from parking meters
  • Architected & built a blockchain-based proof-of-concept in the supply chain domain by identifying the pain points of the source-to-pay process flow using Ethereum platform with NodeJS backend and MySql database
  • Developed visualization backend using NodeJS and MongoDB to facilitate stakeholder specific view of the network

Education
Duke University
2019-2021
MS in Data Science
CGPA: 3.72 out of 4
Taken Courses
  • Causal Inference
  • Deep Learning
  • Natural Language Processing
  • High Dimensional Data Analysis
  • Machine Learning
  • Data Analysis at Scale in Cloud
  • Data Visualization
  • Data Science Ethics
Indian Institute of Technology, Bombay
2015-2017
M.Tech in Communication and Signal Processing
GPA: 4 out of 4
Publications
Taken Courses
  • Signal Processing
  • Linear Algebra
  • Information theory
  • Embedded systems
Visvesvaraya National Institute of Technology
2011-2015
B.Tech in Electronics and Communication
GPA: 4 out of 4

Projects

Card image cap
Deep learning Approach for Question and Answering

Introduction Question answering is one of the most challenging applications we have in Natural Language Processing. It is widely used in applications like information retrieval and entity extraction. It is also used as the back-end framework for systems like chatbots and for simulating human-like conversations. While it is ambitious to create and evaluate the performance of an agent in general dialogue, it is relatively easier to evaluate its responses to input questions.

November 10, 2020 Read
Card image cap
Facebook Political Ad analysis

This project explores Facebook ads run by the Presidential candidates and Congressional candidates in recent US elections. Using Natural Language Processing techniques we analyze the content, tone, and targets of digital communications in an effort to better understand the kinds of messages that candidates highlight to voters online. Project Vision Observers have speculated that political ads on social media contribute to growing divisions in American society, engaging in partisan mudslinging and uncivil attacks over respectful discussion of the issues.

August 1, 2020 Read
Card image cap
Matrix Factorization methods for Recommendor Systems

This project compares two deep learning methods used in recommendation systems and compares the results of these methods on MovieLens dataset. Abstract Recommender systems are widely used tools to make personalized recommendations to individuals. Neighborhood and factor models have been extremely popular techniques in the field of recommender systems. However, given the successful application of neural networkbased techniques in other fields of technology such as computer vision and natural language processing, it is reasonable to believe that neural network-based techniques would also perform well in recommender systems.

March 1, 2020 Read

Image captioning
Image captioning
Contributor March 2020 - April 2020

Developed a web-app which takes an image as input and generates a caption. The project uses a CNN encoder & RNN decoder to train the model for caption prediction.

Real time twitter sentiment analysis with AWS
Real time twitter sentiment analysis with AWS
Developer Feb 2020- Apr 2020

Developed a serverless data engineering pipeline to download tweets from users and analyze the tweet sentiment using AWS comprehend API.

News recommendor system
News recommendor system
Data Scientist at Capital Group Jun 2020 - Aug 2020

Developed a recommendation model that uses user and news metadata and recommends top 5 articles leading to 50% higher click rate

Survey Data Analysis
Survey Data Analysis
Data Analyst Jan 2021 - Present

Analyzing data collected by Duke Graduate school from research scholars to help improve department engagement.

Toxic comment classification
Toxic comment classification
Developer Apr 2020- May 2020

This project aimed to detect different types of of toxicity like threats, obscenity, insults, and identity-based hate on online platforms using Naive-Bayes- SVM and LSTM based model.

Detection of solar PV in aerial imagery
Detection of solar PV in aerial imagery
Developer Feb 2020- Apr 2020

This project aimed at identifying images with solar photovoltaics in images captured from satellites.

Flask app to visualize stock data
Flask app to visualize stock data
Developer Mar 2020- Apr 2020

Developed an app to fetch stock data from Yahoo Finance API for the given company, plot the opening and closing prices for the company’s stocks over the last three months.

Spotify Song Popularity Prediction
Spotify Song Popularity Prediction
Developer Nov 2019 - Dec 2019

The project aimed to understand the features that hit songs have in common using data downloaded from Spotify API.

Opportunistic scheduling in two-way wireless communication with energy harvesting
Contributor Jul 2016 - Jul 2017

In this paper a two-way half-duplex communication model is considered and problem of minimizing the sum of the time required to send the required bits in both the directions is solved.

Details
DInEMMo - Decentralized Incentivization for Enterprise Marketplace Models
Contributor Aug 2018 - Oct 2018

In this paper, we presented DInEMMo, a solution that is built on the convergence of decentralized AI and Blockchain. DInEMMo is enabled with configurable smart contracts with added features.

Details

Skills