Close

Divyanshu Jhawar

 

Data Scientist

Download Resume

About Me

A traveller by heart, learner by birth and programmer since a long. I am data-science enthusiast and love to work on neural nets. I like to explore all around and travelling has been my holiday interest. I am a fan of multiple sports: cricket, badminton, soccer, tennis, F1 (though I don't play all 😜). In spend my free time on video games or photography.
It is due to my course-work, projects and internships, that I was familiarized with wide variety of computer science concepts. My work and experience is related to data science as well as software engineering involving web development and android development.

Experience

Emoty.AI

Data Science Intern

  • Led the project and designed an end-to-end pipeline for training deep learning models and running inference on AWS EC2.
  • Successfully deployed a multi Convolutional Neural network for human eye gaze detection regression model. Leveraged optimization and data transformation techniques which reduced 30% running time.
  • Managed image data-set on AWS s3 with a size of 150 gigabytes, including over 2.4 million recordings.
  • Performed 16x data augmentation to increase the model robustness using OpenCV. Worked with GANs to generate new facial expressions and robustness.

O'Neill School of Public and Environmental Affairs

Data Analyst (Research Assistant)

  • US county-level analysis, on how&why people are moving, the percentage change in movement within and outside the county. Creating data pipelines using python scripts based on the 6 months data combining over 50 million rows.
  • Created a data-mining pipeline on AWS Snowflake and used SQL to work with 100 million row datasets.
  • Managed image data-set on AWS s3 with a size of 150 gigabytes, including over 2.4 million recordings.
  • Performed visualizations on GIS data and shapefiles to analyze the sentiments behind the movement in each region, using Plotly Mapbox, Express, Graph-objects, Time-Series, and ScatterGO.

EnClient Pvt Ltd

Software Trainee Intern

  • Developed a speech recognition system for the android app to enhance auto-answering. Successfully integrated voiceSpeechRecognizer API in the android application which has resulted in faster walk-through for users.
  • Implemented Alexa Voice Service(AVS), which handles in-time translation functions and analytics on the AWS cloud.
  • Using the Lexbot facility of AWS created a Lambda function to deliver chat-bot functionality. This enables auto-request submission by the end-user.

Universiti Putra Malaysia

Machine Learning Research Intern

  • LSTM RNN architecture was used for the classification of human’s day-to-day activities. IoT based ARAS binary sensor data-set was chosen for the purpose that it is recorded in the natural environment and had over 2.2million entries. Adam optimizer was used due to its capability to handle sparse data and also maintains the average gradient.
  • Apart from data pre-processing, it involved reviewing various other papers that have been published in the same area. This problem was a multi-class multi-label classification. Mini-batch LSTM and Deep LSTM approaches were used for activity prediction. Combined accuracy on the data-set came out to be 72.2% and 77.7% using both the approaches. The work will be published in SmartCom 2020 Springer Edition.
  • Both the models were implemented in Tensorflow with a mini-batch of size 128. The dropout layer was added to reduce the over-fitting present in the model. The model was run on Nvidia Quadro K4200 GPU.
  • Paper is published in the International Conference on Paradigms of Computing, Communication, and Data Sciences.

National Ilan University

Artificial Intelligence and Multimedia Intern

  • My work as a researcher was to work on the stability of natural bio-resources that generates electricity. The aim was to predict the cycle, after which the electricity generation becomes stable and the system can be stopped from there.
  • Our research had produced two novel algorithms to accomplish the task. The first algorithm was based on the similarity index between two cycles, and the second involved smoothing of the moving average. Interpolation techniques were used to generate converging cycles.
  • Published Journal Paper can be found here- “An Efficient Computational Model for Assessing the Stability Characteristics of Electro-active Natural Bio-resources”, Recent Patents on Computer Science (2019) doi.org/10.2174/2213275912666190809120031
  • Not only this my stay at NIU was accompanied by analyzing the working of applications based on Virtual and Augmented Reality, like VR games, animations.

Hackathons

Cathacks VI Hackathon

May 2020

 

LNM Hacks 2.0

Nov 2018

 

Education

Indiana University

Aug 2019 - May 2021

Master of Science in Data Science

GPA: 3.89/4

The decision to pursue MS was solely due to the fact that I wanted to dive deeper in the field of Data Science. My learning has been a combination of diverse set of courses in Data Science. It involves working with bayesian learning, Aritficial Intelligence, statistical modelling, big data, and their practical applications. Coursework:

 

The LNM Institute of Information Technology

Aug 2015 - May 2019

Bachelor of Technology in Computer Science

GPA: 7.6/10

My Bachelors' journey had been an uphill drive, I started low but came out strong and ended up satisfying myself of what I had aspired to be. I had earlier worked on with Android applications, for around 3 semesters. I got expertise in algorithms and data-bases. Then I was inspired very much by the recent advancements in the AI, so decided to try on with Machine Learning. From 6th to 8th Semester I chose various subjects related to Data Science:

 

Projects

Eduform- Learning Management WebAPP

An AWS hosted web-app that provides a platform to connect learners and educators. This app provides a one-place solution to the learning management, including, user auth, Googele auth, user verification, learning, grading, assignment submissions/download, customized theme, and chat.

View Project

PySpark- Book-Crossing

A comprehensive analysis of Pyspark and SQL on Book-Crossing Dataset, involving the extraction, transformation, loading, processing and visualization. The project is built on Virtual Machine set up on JetStream.

View Project

Artificial Intelligence Games

This repo consists of all my work related to EAI course. It demonstrates various different application of search, abstraction and other algorithms. It has maze solver, 15-tile slider, 2048 AI game and decryption using metropolis hastings algorithm.

View Project

Housing Value Prediction

This project presents a detail comparative study of ML algorithms on California Housing data-set. Algorithms used: Linear Regression, Random Forrest, Ridge Regression, Neural Network and Gradient Boosting. The Gradient boosting performed best due to the fact that the features were mostly weak. So, using grid search and cross validation we were able to get the best parameters for our data-set.

View Project

Facial Segmentation

A demonstration of Segnet architecture based on CNN, to show image segmentation. The project was based on “Labelled Faces in the Wild” where the task is to segment the face away from rest of environment, reaching to an accuracy of around 90%.

View Project

Sentiment Analysis

This work was a part of Kaggle notebook that displays the complete understanding of Bayesian treatment in Naive Bayes Algorithm with K-Fold cross validation. It was performed on three different sentiment data-sets, from Amazon, Yelp and Imdb. The mean accuracy reached to over 80%. Both the maximum likelihood and MAP were compared to get an understanding of importance of MAP solution.

View Project

Time Series Analysis on HAR

This project displays my hands on with Time-Series concept, my attempt to generate the inference in the data through Time Series Analysis. The work as a detailed analysis put up, while attempting various modifications. It displays extensive feature engineering and use of ARIMA model.

View Project

Fruit Recognition Application

This project is an android based fruit classification app using Convolutional Neural Nets on “Fruits-360” data set. The model reached good accuracy of around 93%. I tried to use the trained model in the Andorid environment, but had faced difficulty in using Tensorflow in Android. So, I used open sourced server to run my model at.

View Project

Vehicle Tracking App

Android based application which connects passenger to their bus. This app is divided into two parts, one each for user and bus driver, similarly to what different cab services use. This application was created to smoothen the bus service which is used my many organizations all over the world.

View Project

Hospital Management Application

This project is Android based Hospital management system, which was created for the use of LNMIIT doctors. The app has multiple functionalities, it can update the stock available using QR code, generate the report based on the day, list of stock on a particular day and report is sent directly sent to email of the user. The app was deployed on Firebase, and auth and db is maintained there. The app displays use of design patterns as well.

View Project

Skills

Get in Touch