Artificial Intelligence & Machine Learning

• Designed and implemented the data architecture for Netflix's entire member experimentation ecosystem from the ground up.• Devised novel solutions for delivering daily or incremental refreshes across every experiment (thousands of tests at petabyte scale) that was previously not possible at Netflix due to scale along with an 85% cost reduction.• Implemented a custom materialized view solution with an automated janitor in order to speed up data retrieval (~5000% median) while purging the data when it is no longer needed.• Implemented one of the largest Flink pipelines at Netflix that processes over 2 million records per second to log all treatment exposures for experimentation.

• Lead architect of Facebook's client-side logging design and implementation within Ads Interfaces.• Lead engineer of Facebook's Data Profiler automation tool used across Data Engineering.• Created and led the Data Quality course for Facebook's bootcamp.• Wrote the Code Quality and Engineering Rigor guidelines and principles for the Data Engineering org.• Created and led the Spark Scala course across the Data Engineering org.

• Designed petabyte scale daily ETL using Pig, Python and Java for Netflix’s AB Experimentation Platform• Designed largest ETL at Netflix for aggregation of client-side logging using MapReduce, Spark & Scala.• Created Netflix’s nonmember acquisition dashboard from end to end using Pig, Druid, Javascript and Python.• Built ETL pipelines and dashboards to track Application Performance Metrics across the Netflix service. This was instrumental in guiding business decisions and prioritization of resources.• Designed integration testing framework for the AB Experimentation UI with Python, Javascript, Selenium & Presto.• Built ETL in Spark to load test and evaluate Spark’s effectiveness with Netflix’s data volume.• Built several UDFs for Netflix Pig ETL Library (Python, Java) and Spark ETL Library (Scala).• Built anomaly detection into ETL auditing tools to strengthen data quality.• Developed client side logging specifications for UI Engineers in order to track interactions between all apps that host the Netflix service.• One of the primary contributors to the Pig, Hive and Spark internal Netflix user groups.• Taught best practices for ETL optimization across the organization which translated to large cost savings.• Served as the primary analyst to help identify areas of friction and improve the overall nonmember experience.• Provided insights to inform business on which AB Tests should be pursued and how to interpret client side logging.• Developed algorithms and analytical studies to inform the business on the impact of account and profile sharing.• Wrote several memos used to guide Netflix’s business strategy in India and other less developed markets.• Identified several device usage patterns that impacted engagement and retention.• Worked extensively with Hive & Presto within Python & R for analytical studies and to build Tableau dashboards.• Performed clustering analysis to determine which cohorts of users and payment methods are more likely to fail.

• Designed the logic and algorithms powering Chegg’s current scientific pricing engine.• Built a web crawling framework by utilizing Hadoop Streaming, Tor, Polipo, Python, PhantomJS/CasperJS.• Mined and analyzed millions of public profiles from the web to design a graph based on school, skills and profession through Hadoop Streaming, Pig and Python.• Developed a natural language processing/text classification engine using Python’s nltk, numpy, scipy, and sklearn.• Created and designed Chegg’s Analytics Dashboard using d3.js, jQuery, Javascript, CSS, HTML, MySQL and PHP.• Created all of the MySQL queries that ran on the PHP backend as well as all of the ETL to create intermediary tables to optimize the performance of the AJAX calls used in the Analytics Dashboard.• Designed and modeled Chegg’s textbook forecasting model used for sourcing and pricing using R.• Designed an algorithm to link and cluster textbook editions together for purposes of title economics.• Designed an algorithm to perform matrix multiplication of massively large matrices in MapReduce and Python to perform the TF-IDF/similarity calculations for the Chegg Study text classification engine.• Developed and maintained ETL chains for databases used by the Data Science team.• Created Chegg’s spam filtering engine using TF-IDF and a Naïve Bayes classification engine in Python.• Worked on an algorithm for Chegg’s Career Center product by creating linkages between careers/internships to skill groups and courses to skills in Python.• Designed and deployed Chegg’s current textbook liquidation and buyback pricing engine using R.

• Developed a predictive model with R and SQL that incorporates multidimensional data from Facebook Insights and the United States Census Bureau to determine optimal targeting variables for improving ROI.• Designed a Generalized Linear Model to optimize Facebook promoted post creation and increase its virality.• Created scripts in R to generate automated daily performance reports for customers and the Data Analyst team.• Created data visualization scripts in R to keep track of the performance of Facebook advertising campaigns.• Managed 70% of the GraphScience revenue stream by launching and optimizing Facebook campaigns for clients.• Designed an optimization algorithm in order to build an automatic budget allocation tool that would update Facebook advertisement campaign bids and budgets in real time when certain conditions are met.• Developed data mining scripts in R to scrape web pages and gather key information about the client user base.• Performed routine Quality Assurance checks for engineers on weekly product releases.• Led weekly client calls to discuss performance goals and strategies while alleviating any current and future concerns.

 Ran the Businessowners review to determine prospective loss costs to be implemented in each state and territory. Executed the BOP ratemaking system using SAS to calculate prospective rate changes based on large data sets. Pulled records from large databases with SQL to conduct data adequacy investigations within Access and Excel. Applied Bayesian credibility analysis and Tweedie Family Generalized Linear Models to calculate class plan relativities. Presented internal/external trend analyses and company investigation recommendations to senior management. Trained and mentored staff on BOP data analysis and filing procedures.

 Worked with a team of actuaries and statisticians to create the Risk Analyzer Personal Auto (RAPA) predictive model. Employed various modeling techniques such as spatial smoothing, multiple regression analysis, k-means clustering principal component analysis , cross-validation, and bootstrap techniques. Coded programs using SAS and SQL to parse through large data sets that were used in the rating variables. Built maps and analyzed geographical data used in RAPA through geographic information software, ArcView 9. Created presentations used to explain the RAPA model to Department of Insurance regulators for approval.

 Worked extensively with ISO’s loss and premium database in order to investigate and determine any anomalies reported incorrectly through the Businessowners statistical reporting plan.  Contacted companies to assist them in reporting premiums and losses correctly under their statistical plan. Programmed in mainframe SAS to organize insurance records used for trend and loss cost analysis. Created Businessowners loss cost revision filings for each state to be approved by their respective state regulator.