Bennett Marsh Email & Phone Number

Austin, Texas, United States

Austin, Texas, United States

Austin, Texas, United States

Austin, Tx

• Built WatchCBB (http://watchcbb.me), a web app that provides personalized recommendations of college basketball games to watch• Scraped historical data of 60,000 college basketball games and used this to train a series of regression models to predict game-level features such as score margin, pace of play, and upset probability, achieving performance as good or better than leading sports analytics providers• Leveraged logs of thousands of game threads collected from Reddit in order to forecast the popularity and entertainment value of a given basketball game• Constructed and maintained a daily-updating PostgreSQL database with cumulative season statistics for each team, upcoming schedule information, and the latest Reddit game threads• Designed and deployed a web app with flask on AWS, which queries data from the PostgreSQL database in order to provide customizable recommendations of games to watch in the coming week

Santa Barbara, Ca

• Processed, validated, and analyzed tens of billions of events of proton collision data in a search for new sub-atomic phenomena. Lead researcher on the inclusive MT2 analysis, one of the first at CMS to publish on the full 137 fb⁻¹ of Run 2 data• Managed large-scale distributed computing workflows for parallelized processing and analysis of petabyte-scale experimental datasets• Led a team in the development and deployment of machine learning-based anomaly detection tools to automate a time-consuming data quality monitoring workflow, eliminating 80% of required manual labor. Utilized both supervised and unsupervised methods to detect anomalies and flag problematic behavior• Trained a series of boosted decision trees and neural networks to discriminate signal from background in a search for rare Higgs boson decays, improving the signal-to-background ratio by over 40%• Designed and managed a multi-step data-processing pipeline for a new experiment searching for dark matter, including event generation, detector simulation, data cleaning, and final analysis• Communicated results via frequent oral presentations and written internal documents

Geneva, Switzerland

• Configured the CMS hadronic calorimeter's online software to run under valgrind and developed computational tools to analyze and extract useful information from the memory-usage log files• Studied the source code to identify the sources of memory leaks and suggested ways to eliminate them

West Lafayette, In

• Implemented genetic algorithms to optimize an empirical term meant to correct DFT energies for Purdue's Theoretical & Computational Biomolecular Physics Group• Extended the methodologies to other non-DFT energy calculation schemes

Houston, Texas

• Designed an interactive tool to visualize coordinate frames in International Space Station simulations, streamlining the process of developing and debugging the simulations• Built a Tcl/Tk plugin to scatter rocks across virtual planets, eliminating the need for much tedious design work

West Lafayette, In

• Investigated the use of high-throughput computational methods on a 3D image reconstruction tool on Purdue's Diagrid computing network• Implemented parallel job submission capabilities to allow support for large datasets

Ph.D., M.A., Physics

Dissertation: "Search for new physics using the MT2 variable in all-hadronic final states produced in 13 TeV proton-proton collisions.

B.Sc., Honors Physics, Mathematics