Software Engineer with over 8 years of experience specializing in Backend Engineering, SRE and DevOps. Ex-DigitalOcean

Anand Vyas  • Peerlist

Implemented the Distance Vector Routing and Distance Vector Multicast Routing Protocol in Java on Unix Systems. Individual processes for Host, Router and Controller were implemented.

Advanced Computer Networks - Distance Vector 

Developed an automated test prioritization tool that can be used with any java project to prioritize unit test execution in order to discover bugs faster. Technologies used: Java, Maven, JUnit, Java Agent, ASM Bytecode Manipulation Framework Without manipulating or even knowing the structure of the source code or unit tests, this tool can be deployed on any project to create a new - prioritized test suite. Say you have a 100,000 unit tests in your project and it takes a few hours for all the unit tests to run. Now, if  there was a strategy that could run these unit tests in a specific order of priority, so that you could discover bugs faster wouldn't it be very helpful and save you tremendous amount of time in discovering bugs? The answer is yes. The strategy we have implemented to achieve this is Automated Test Prioritization using statement coverage information of unit tests. Statement Coverage: Number of (unique) source code lines that are covered by a unit test. The Test Prioritization is based on two strategies: Total and Additional strategies. Total strategy - The unit tests with highest statement coverage are prioritized first. Additional strategy - The unit tests which cover maximum number of 'yet uncovered' source code statements, by previous unit tests,  are prioritized first. The experimental data shows that these strategies are able to detect bugs faster rather than randomly executing the test suite. Logically, a unit test covering more lines will have a higher priority of detecting bugs than a test covering few lines.

Automated Test Suite Prioritization Tool

I was responsible for performing requirements collection, developing ER/EER diagrams, perform relational mapping and normalization (up to 3NF and BCNF), write stored procedures using PL/SQL. Technologies: SQL, PL/SQL, Oracle SQL Developer, ERDPlus Software

Database Design for Taxi Management System So

The Fixed Income Decision Support System is State Street Global Advisors', in-house Portfolio Construction Software made as per the specific requirements of its Portfolio Managers. Principles of Agile Software Development were followed in all phases of development. We followed the SCRUM framework rigorously. The team was extremely knowledgeable on the business and technical arenas and all the members of the team have given me priceless mentorship for which I am ever grateful. It is a thick client application. We developed the front-end using WPF and C#. Net. Service layer was developed using Java and backend databases were from Oracle. We wrote stored procedures in PL/ SQL. I also learnt about version control software, continuos delivery and extreme perfection for quality. We followed test-driven development and wrote unit tests for each piece of code added for new stories. Integration tests, regression tests, smoke tests and tests for the application at each stage were rigorously performed to ensure quality. [Levels: Development -> Development Integration -> User Acceptance Testing -> Production] Invaluable exposure to the Financial Services industry. Thanks to all these amazing people and many more in associated teams.

Fixed Income Decision Support System

Combined two disparate datasets - airline data and weather data, each containing millions of entries, to show that weather is a significant predictor of delays in flights. This was done by showing improvements in prediction accuracies before and after combining the weather data. Used Scala, Spark and SparkMLLib machine learning libraries. Machine learning algorithms used to calculate prediction accuracy: Logistic Regression, SVM, Decision Tree and Random Forest

Flight delay prediction using Scala, Spark an

A Decision Tree is a Machine Learning concept used for classification of new data. The tree is created by a concept called 'training'. After training (and pruning), new data can be entered as input, and we can get a classification as output. This is different from regression where predictions are made. An accuracy of 70-75% is considered to be good. Implemented a decision tree using the ID3 algorithm using the Information Gain (Entropy) Heuristic and the Variance Impurity Heuristic. Coding was performed using OOP concepts in Python. Also, implemented the post pruning algorithm on the decision tree to improve its classification accuracy. Received a good accuracy of 75% on the test data set after application of post pruning algorithm.

Machine Learning - Decision Tree Implementati

Implemented the Perceptron Training Rule in Machine Learning with 90 + percent accuracy on filtering Spam Emails.

Machine Learning - Implementation of the Perc

Implemented the K Means Clustering Algorithm for Image Segmentation. K Means is an 'Un-Supervised' learning algorithm in Machine Learning.

Machine Learning - K Means Clustering for Ima

This was a large scale, semester long project. We used the Scrum Framework for developing this project. Python was the language of our choice and the entire project has been built from scratch without using any tools/ API's. We implemented the Collaborative Filtering Algorithm using the User - User Approach. We have used the Pearson Similarity Coefficient, for measuring the similarity between users. We used an open source Book Dataset (CSV files) and wrote a Python Script to extract the required data into a particular format. Despite of the large dataset, we were able to achieve a small Root Mean Square Error on the accuracy of predicted ratings. We measured both one and multifold accuracies. We also researched extensively on the methods used by the winning team of the NetFlix Prize. We researched on and presented the following algorithms in class: Normalization of Global Effects, Neighborhood Models, Implicit Data, Matrix Factorization, Regression, Restricted Botlzmann Machines, Temporal Effects, Regularization and Ensemble Methods.

Machine Learning - Recommendation Engine usin

Using the Multinomial Naive Bayes algorithm and the Logistic regression algorithm I was successfully able to train the program such that the spam and ham (non-spam) messages were filtered with 92-95% accuracy on the test data. The codes for both the algorithms were optimized and amongst the fastest to give high accuracy outputs.

Machine Learning - Spam filtering using Multi

Developed projects based on Monte Carlo Simulations, Point and Interval Estimation, Hypothesis Testing, Chi-Square Tests, Non – Parametric Tests, Regression and ANOVA. Technologies: R Programming Language, R Studio.

Mini Projects: Statistical Methods for Data S

Developed a professional social networking platform for members of NMIMS University specifically designed for information sharing. Comprehensive documentation of the project was also done by the team. Technologies: HTML5, CSS3, Bootstrap, JavaScript, PHP.

Professional Networking Website: College Comm

My responsibilities were as enlisted below: Write programs in SPARQL to get aggregated data from RDF/XML files (2009 Veterans Health Administration Report) Generate statistical visualizations from these data in the form of different types of charts using the Google Visualization API. Develop a website to showcase these visualizations. Technologies: SPARQL, Google Visualization API, Apache Jena Fuseki Server, RDF/XML, HTML5, CSS3, JavaScript, AJAX, Bootstrap, JSON.

Semantic Web: Statistical Visualizations for 

Developed a SDN Firewall Application using the POX (Python Based SDN Controller) and Mininet. Scripts for the Northbound Application were written in Python.