Play Games24x7 Pvt. Ltd.

Software Development Engineer, RummyCircle

---

• Helped build the backend platform which allows massive amount of users to play online card-based games such as Rummy.


• Contributions include building, testing and deploying rest-based Web Services, providing application-level security to these services, maintaining bots to test gameplay.

Skills: Java, Spring, Test-NG, Hbase, Ehcache, Redis, MySql

Dates: July 2013 - May 2016

Close

KDE (Google Summer of Code)

Software Developer, Choqok

---

• Integrated features of Facebook into KDE's micro blogging client, Choqok (Under Google Summer of Code 2012).


• Built a base library which allows users to create, view, like and comment on posts

Skills: KDE4 based on Qt (C++ based GUI Framework).

View Project on: GitHub

Dates: May-July 2011 & May – August 2012

Close

Oracle

Automation Engineer

---

Worked as an Automation Engineer with Merchandise Operations Management (MOM) Team at Oracle Retail Used technologies like HP Quick Test Professional and Selenium Webdriver for scripting and execution of test cases.

Technologies: Java, Selenium WebDriver, HP Quick Test Professional, VBScript

Dates: December 2012 - May 2013

Close

CiBO Technologies

Data Scientist

---

• Lead the development of a Bayesian statistical model to automatically identify the crop grown in an agricultural field along with the planting and harvest date, using remotely sensed satellite imagery.
  • Used Monte Carlo sampling methods to learn best-fit parameters and obtain predictive distributions.
  • Scaled the model to efficiently generate predictions for fields across the contiguous United States.
  • My work here has contributed to 2 pending patents, and we are also in the process of drafting a publication to one of the remote sensing journals.


• Built a model to interpolate cloudy and missing pixels in satellite images of agricultural fields.
  • The images generated by this Gaussian Process model give a more complete picture of crop growth in a field, and are useful to build new CIBO products that are potentially robust to missing satellite data.


• Contributed to building a CIBO algorithm that maps the performance variability of a field across multiple years, using satellite imagery.
  • Expanded the existing algorithm to work on additional crops and geographic regions within the US.
  • Designed a numerical score to summarize the maps produced by the algorithm into a single field-level metric.
  • More information about this product can be found in my blog post at How Remotely Sensed Performance Zone Maps Improve Agriculture Assessments


• Processed a variety of geospatial data from various sources to generate statistical and scientific information.
  • The sources include satellite imagery such as Landsat and Sentinel-2, UAV imagery such as Precision Hawk, and other geospatial datasets obtained from USDA, USGS and NASS.
  • Used geospatial tools like QGIS to visually and quantitatively evaluate data.
  • I also have some basic experience working with geospatial python libraries such as shapely and geopandas.

Skills: Bayesian Inference, Data Analysis, Python, Gaussian Process, Scala, Stan, QGIS, Pandas, Numpy, Altair.

Dates: June 2018 - Present

Close

Fermilab (Volunteer position)

Independent Researcher, Deep Skies Group

---

• Employed Bayesian Neural Networks (BNNs) on telescope images to learn parameters for theoretical models in astronomy. These models predict the phenomenon of strong gravitational lensing of sources by foreground galaxies.


• Assessed the usefulness of BNNs by studying output distributions and the resulting uncertainties (epistemic, aleatoric).


• Presented my work as a poster at the 236th meeting of American Astronomical Society, see reference below.

Poster Reference: Bhambhani, P., Hurd, H. and Caldeira, J., 2020, June. BayesCurve: Read your Strong Gravitational Lens with Bayesian Neural Networks. In American Astronomical Society Meeting Abstracts# 236 (Vol. 236, pp. 241-04)

Link to Poster: BayesCurve AAS 236 Poster

Skills: Bayesian Neural Networks, Variational Inference, PyTorch, FastAI, Python

Dates: April 2019 - December 2020

Close

Dark Ecology Project(UMASS Amherst MS Computer Science Research Project)

Research Assistant

---

• Helped build machine learning models for biologists to analyze bird migration patterns, using weather radar imagery.


• Contributions include building a benchmark data set and scaling models to run on large volumes of data on AWS.


• Our work was published at the AAAI 2020 conference. (see reference below)

Paper Reference: Cheng, Z., Gabriel, S., Bhambhani, P., Sheldon, D., Maji, S., Laughlin, A., & Winkler, D. Detecting and Tracking Communal Bird Roosts in Weather Radar Data. AAAI 2020: 378-385 arXiv:2004.12819

Link to Paper: Dark Ecology AAAI 2020

Dark Ecology site: darkecology.cs.umass.edu (WebArchive link as website was down at time of this writing.)

Skills: Data Analysis, MATLAB, Python, Docker, AWS

Dates: May 2017 - May 2018

Close

Environmental effects on Galaxy Population

Astrophysics Msc Thesis, LJMU Astrophysics Research Institute (ARI)

---

• For my thesis project, I am studying the correlation between the number density of galaxies in a cluster and the fraction of galaxies that are red (i.e don’t form new stars), using data from a large-scale astronomical galaxy survey.


• This is a big data project, and involves statistical analysis, model fitting and Monte-Carlo based parameter estimation.

Skills: Data Analysis, Python, Pandas, Numpy, Stan, Altair

Dates: June 2020 - Present

Close

Image Captioning with Visual Attention

Machine Learning Course Project

---

• Incorporated an attention mechanism with gradient descent based training algorithms to build a deep learning model that learns to generate image descriptions given an unseen image, inspired by the state-of-the-art work in this field.


• Our results showed that attention based deep learning models can learn to produce good image captions and generate visualizations to infer how a model can focus on certain parts of the image while predicting the next word.


• We used a Convolutional Neural Network (CNN) to generate attention based features and a Long Short-term Memory Network (LSTM) for generating image captions.

Skills: Keras, Theano, Python

View Project on: GitHub

Dates: October - December 2016

Close

Automated Extractive Summarization of a Book using its Index

Natural Language Processing Course Project

---

• Generating coherent summaries of long documents such as books is a relatively untouched field, with most of the research work being focused on short documents.


• In this Project, we looked at summarization of books which have a back-index as a novel technique for generating good summaries. We provided a new dataset for the evaluation of such summarization systems. We focused on extractive summarization using LexRank, an unsupervised learning algorithm based on eigenvector sentence centrality and compared it with the current state-of-the-art methods.


• The results were promising and merit future research.

Skills: NLTK, Python

View Project on: GitHub

Dates: October - December 2016

Close

Pictionary with Jibo

HackUMass 2016

---

• Taught the robot Jibo to play Pictionary on a set of hand-drawn images.


• We used Jibo SDK to capture a picture and send it off to be classified by our custom-trained TensorFlow neural networks. We used openCV to detect the playing area within the board and crop off the rest of the image and make it black and white, we then passed it off to our TensorFlow model to be classified.


• We built the neural network using hundreds of images pulled automatically from Bing Image Search. Using TenserFlow's Image Recognition Algorithm "Inception", we successfully trained it to recognize common objects such as house, chair, etc.

Skills: Jibo SDK, Tensorflow, OpenCV, Python

View Project on: GitHub, DevPost

Dates: October 2016

Close

Duckworth Lewis (D/L) Calculator for Android

Personal Project

---

• Designed and Built Duckworth Lewis Calculator - an Android application which lets users predict scores and results in rain interrupted cricket matches.


• Based on the Standard Edition released by the Internation Cricket Council (ICC). Works for domestic as well as international cricket matches.

Skills: Android

View App on: Play Store, GitHub

Dates: April - May 2015

Close

Evaluating Machine Learning Techniques for Website Fingerprinting Attacks

Independent Study with the Secure, Private Internet Research Group (SPIN)

---

• Pursued an independent study to evaluate the assumptions made by different machine learning models for targeting Website Fingerprinting (WF) attacks on the Tor network.


• Existing models include Naïve Bayes, Decision Trees and SVM, but they all employ certain restrictions (e.g., user visits only a small set of web pages, a single tab whose content is unchanging) that aren’t realistically attainable. Evaluated the assumptions made by such models and the performance when these assumptions are violated.


• Proposed two neural network models (a single-layer network as well as a 1D CNN) as a tool to effectively classify multi-tab, open-world, time-variant crawl data, and evaluate their performance against the state-of-the-art SVMs. Results are promising and merit further research.

Skills: Keras, Numpy, Scikit-learn, Python, C++, Bash.

View Project on: GitHub

Dates: May - August 2017

Close

In The Bin!

Google Hacking For Good 2017

---

• In The Bin is a location-based, crowd-sourced web application where users can submit locations of publicly-accessible trash and recycle bins near by.


• Built during the Google Hacking For Good 2017 hackathon at Harvard Innovation Labs.

Skills: Firebase Web, Google Maps Javascript API, Material Design Lite

View Project on: GitHub

View Demo at: Firebase

Date: June 17 2017

Close

Classification of Astronomical Objects from Light Curves

Kaggle Data Science Challenge

---

• Used Convolutional Neural Networks (CNN) to build a classifier that detects a variety of astronomical objects from simulated light curves data provided by the team at Vera C. Rubin Observatory (aka Large Synoptic Survey Telescope).


• Documented my work and published a Kernel explaining my approach - see this Kernel


• I have been invited to present my work at the upcoming LSST Supernova workshop (postponed due to Covid).

Skills: Keras, Sklearn, Pandas, Numpy, Python, Bash

Dates: April 2017

Close

EarthQuake Prediction using Acoustic Time-Series Signals

Kaggle Data Science Challenge

---

• Predict the time remaining to earthquake using acoustic time-series signals simulating seismic waves as generated by the Los Alamos National Loboratory


• Best results obtained using tree-based ensemble models trained on a variety on statistical and signal processing features.

Skills: Python, Pandas, Numpy, Scipy, Keras

Dates: Feb 2019 - Present

Close

University of Massachusetts Amherst

MS - Computer Science

---

Relevant Coursework – Machine Learning, Natural Language Processing, Artificial Intelligence, Systems, Reinforcement Learning, Blockchains

Dates: September 2016 – PRESENT

Close

Dhirubhai Ambani Institute of Information and Communication Technology

B.Tech. - Information & Comm. Technology

---

Relevant Coursework: Algorithms, Data Structures, Computer Networks, Databases, Operating Systems, Systems Software, Cryptography, Security Protocols, Graph Theory, Computer Organisation, Natural Computing

Dates: July 2009 - May 2013

Close

Liverpool John Moores University

MSs - Astrophysics (Distance Learning)

---

Coursework – Introduction to Astrophysics, Observational Astrophysics, Time Domain Astrophysics, Cosmology

Research Thesis – Environmental Effects on Galaxy Population (see Research Experience section)

Dates: September 2019 – PRESENT

Close