HARSH HARWANI

: goo.gl/DRN46A

: goo.gl/VYun3O

: goo.gl/Mnl2HR

EDUCATION
Master of Science, Computer Science
University at Buffalo, State University of New York
Bachelor of Engineering, Information Technology
Gujarat Technological University, India

: hharwani@buffalo.edu

: (716)-697-6596

Graduating-Dec 2015
GPA-3.60
June-2012
GPA-3.50

COURSEWORK
Analysis of Algorithms, Information Retrieval, Distributed Systems, Machine Learning, Data Mining, Data Intensive Computing
TECHNICAL SKILLS
Keywords: OOP, Java, Android, Lucene, Solr, Hadoop, Oracle, MySQL, Web Services, Rest API, spring MVC, JDBC, Maven, Junit.
Languages: Java, JavaScript, Python, SQL, C++, HTML, XML, JavaScript, JSON, CSS, SOAP, Pig, Hive.
Framework and Tools: Struts MVC, Android, Map Reduce, Git, Eclipse, Facebook Graph API, twitter API, MongoDB, Unix/Linux.
ACADEMIC PROJECTS
News Search Engine (J2EE)
Fall-2014
Implemented a News Indexer which contained Parsing, Tokenizing, Filtering and Indexing of about 12000 news articles.
Programmed two popular scoring mechanisms (TF-IDF and Okapi BM25) to rank retrieved documents.
News Personalization (J2EE, JSP, MySQL, Apache Solr, Apache Lucene, Facebook Graph API)
Fall-2014
Developed a news personalization system which captures user behavior and preferences to serve news stories.
Used Apache Solr to index 32000 news articles, used Facebook Graph API and Collaborative Filtering to capture user preferences.
Simple Amazon Dynamo - Replicated Key Value Storage (Android, Java, Multithreading)
Spring-2015
Designed and implemented a simplified version of Amazon Dynamo.
Implemented Quorum replication with replication degree 3 and versioned all objects to differentiate between stale copies.
Incorporated a functionality for handling node failures and ensured synchronization when any node recovers/rejoins.
Stock Volatility Analysis (Map/Reduce, Hadoop, Hive, Pig, Java)
Spring-2015
Developed a Map-Reduce application which was able to analyze large stock data sets of size 1GB and about 40000 files.
3 Map-Reduce jobs were chained together to get most volatile stocks and performance was analyzed on various sizes of data sets.
Same task was done using pig and hive, approximately a 101% improvement in the execution time was observed.
Predicting Diabetes Level in patients (Python, NumPy, Matplotlib)
Spring-2015
Predicted diabetes level in patients using four machine learning techniques a) Linear Regression, b) Ridge Regression,
c) Ridge Regression using Gradient descent, d) Non-Linear Regression and compared their performance.
Used the results obtained to make recommendations for anyone using regression for predicting diabetes level.
Handwritten Digits Classification (Python, NumPy)
Spring-2015
Classification of dataset of handwritten digits consisting 60,000 training samples through Multilayer Perceptron.
Implemented a Neural Network (MLP NN) and back propagation algorithm incorporating regularization, achieved accuracy ~ 92%
WORK EXPERIENCE
Tata Consultancy Services, Ahmedabad, India
Assistant Systems Engineer
Dec 2012-July-2014
Value Added Tax Application (J2EE, JavaScript, Struts, BC4J, Oracle, Jasper Reports, Junit, Maven)
Collaborated across multiple teams through the entire lifecycle to develop various modules using J2EE, JSP, Servlets, Struts, Oracle.
Developed various modules for a Value Added Tax collection application of a state government.
Designed and programmed an FIFO (First in First Out) approach for the task approval flow, this change reduced corruption, also
reduced the task approval time by about 10-12%.
Hospital Management System (C++, Oracle)
Designed and Developed a Hospital Management System using C++ in the front end and Oracle as the backend.
Received appreciation for the attention to details and good leadership qualities.
INDEPENDENT PROJECTS
Classic Arcade Games (Python)
SUMMER-2015
Implemented various classic arcade games like pong, memory, black-jack and snake.
Algorithms used in these games for collision detection, friction, sprite animation were written from scratch.
Twitter Sentiment Analysis (Python, NLTK Toolkit)
SUMMER-2015
Trained a Naive-Bayes classifier using 50000 hand classified tweets to predict sentiment of a tweet as positive, negative or neutral.
Collected last seven days data of a keyword given by the user using Twitter REST API, cleaned the data and fed it to the classifier.
Achieved an accuracy of about 75% for the classifier, presented the results using the Google charts API.