Neil Chainani

• Performed basic log file analysis using Apache Spark’s map-reduce functions on Federal Trade Commission’s fraud and complaint reporting website.
• Evaluated various JavaScript libraries for fraud and complaint visualization.
• Designed mapping interface using Mapbox as a means of geographic filtration of search results.

• Implemented D3.js, a data visualization library, to program an interactive visualization dashboard to aid client in making data-driven decisions.
• Researched and selected the most effective data visualization based on assessment of client needs.
• Worked in a team to present recommendation and dashboard to client.

Intensive 1-year master's degree covering statistics and computer science.

Coursework:

AC209 Data Science \\

CS181 Machine Learning \\

AM207 Stochastic Methods for Data Analysis \\

ST139 Statistical Sleuthing for Linear Models \\

CS205 Computing Foundations for Computational Science \\

CS207 Systems Development for Computational Science \\

AC290R Extreme Computing \\

AC297R Capstone

Minor in Statistics

• Gathered data from nearly 7,000 companies using the Crunchbase API to build models to predict the success of start-ups using an ensemble of six different machine learning methods.
• Built a similarity model to cluster certain start-ups together using a distance measure, and visualized these connections in a dynamic node graph.
• Designed a website to summarize findings and provide interactive D3 visualization of results.

http://nicodri.github.io/CS109_crunchbase

• Used Apache Spark to design from scratch two machine learning algorithms, ordinal regression and random forests, applying parallelization techniques for optimal performance across distributed nodes.
• Benchmarked performance against existing libraries in Spark and scikitlearn to determine amount of speed-up achieved.

http://abhishekmalali.github.io/spark-ml

• Performed simulated annealing, greedy algorithm, and other heuristics to select optimal users which maximized influence over Yelp's social network.
• Modeled influence of a node using a stochastic independent cascade function
• Parallelized computation in Apache Spark and calculated uncertainties of estimates.

In this project, we contrasted two different questioning schemes to assess efficacy in recovering true class labels from noisy labels provided by crowdsourced non-experts, when there are more than two classes to choose from. We generate data with a confusion matrix for each expert, and an underlying class distribution, and attempt to recover both parameters and the true labels. We used Expectation Maximization (EM), Simulated Annealing, and PyMC to compare efficiency and verify results.

This is a project for the capstone course, where I worked with students from Harvard and Politecnico di Milano to create a web application that connected companies looking to create revolutionary new products with the design community. We designed the product from idea to implementation, contacted real companies, and tested out the platform for a data visualization class. I personally was in charge of the django-based back-end.