Data Scientist, System1, Venice CA
April 2017 - Present
- Build and maintain RPC and CTR models to maximize revenue per impression
- Implemented a Bayesian Thompson Sampling Bandit to dynamically compare and select algorithms on live traffic
- These optimization products typically provide 14-18% lift per day (on a base revenue of 500-600K each day)
- Tools and Tech: Python, R, Redshift, S3, EC2, Docker, RStudio, Jupyter
Consultant, OnPrem Solution Partners, LLC, Los Angeles CA
Sept 2015 - March 2018
WB Technology Solutions, Data Analysis and Integration Support
Aug 2016 - Nov 2016
- Wrote Python scripts to normalize and match titles from distinct systems allowing film/fan data to be linked with ticket sale data
- Helped create, update and standardize data documentation and ERDs
- Tools and Tech: Redshift, S3, ER/Studio, Python, Google Cloud Platform, Apache Hive/Hadoop
Disney, DELTA Application Reporting
June 2016 - Aug 2016
- Built dashboards to provide insight into application performance leading to faster bug detection and order completion
- Tools and Tech: MySQL, Tableau, Bash
NBCU Studio Operations ORBIT Billing Phase 1.5
Oct 2015 - April 2016
- Automated data exploration and validation procedures using Python, R and Dynamic SQL
- Performed data modeling and wrote reports for an interim reporting system
- Tools and Tech: Python, CAErwin, DB2, Teradata, R, Batch Scripts
Graduate Student and Researcher, University of California, Los Angeles, CA
June 2009 - March 2015
Completed the Degree of Doctor of Philosophy in Psychology
Jan 2013 - March 2015
- Developed the statistical theory and computational algorithms needed to address a known multivariate modeling issue
- Analyzedtheperformanceofthetheory/algorithmsinsimulationsandappliedthemtorealdata
- Presented results to a committee of statisticians and psychologists
- Tools and Tech: R, EQS
Graduate Student Research Mentorship Award Recipient
June - Sept 2011
- Awarded funding to work on a machine learning problem in neuroimaging
- Results determined which of several methods could most accurately estimate the number of independent neurological systems contributing to the (highly noisy) fMRI signal
- Tools and Tech: Matlab
Graduate Research Methods Project
June 2009 - Dec 2010
- Implemented and applied several heuristic optimizers (e.g. genetic algorithms, particle swarm optimizers, etc.) to select features and weights in a constrained linear modeling task
- Results identified the best performing algorithms and circumstances in which the constrained models might be preferred over the usual unconstrained models
- Tools and Tech: Matlab
Cognitive Neuroscience Research Assistant, Clarkson University, Potsdam, NY
Aug 2008 - May 2009
- Used neural networks to model event-related EEG data to study neurological processing of tactile stimuli
- Tools and Tech: Matlab