Leondra Gonzalez

Led, coached, and developed Wpromotes high-performing Data Analysis team for 12 for multi-million-dollar retail, and entertainment accounts including NBC Peacock, HBO Max, CNN, Tubi (Fox), and DC, Whirlpool, and Coffee Bean, across numerous online & offline marketing mix.
Used Python and R to develop deep learning (RNN), causal inference, game theory (i.e.: Shapley values), Bayesian structural time series analysis, ensemble method forecasting, and optimization for various marketing science applications including propensity modeling, predictive customer lifetime value analysis, media mix modeling, and attribution in AWS and GCP.
Partnered with paid media managers, Strategy, Client Services, Data Science and Engineering to produce a portfolio of analytic solutions and develop efficient workflows that cut delivery time by 50%.

Managed, trained, and developed the agencys central analytics, data science, and data engineering team for global accounts including Procter & Gamble (Febreze), The Art of Shaving, and First Aid Beauty.
Designed statistical models in R to develop landing page regression models for unique ad campaigns, budgets, and placements to optimize campaign performance and RoAS.
Partnered with Ad Ops and client business intelligence (Datorama, Tableau, DCM, Google Analytics, Trade Desk) requirements powered by Python ETL pipelines and dashboard enhancements using custom Java calculated fields.

Managed analytics and process improvement efforts to reduce the cost of multi-million-dollar production and media budgets for Toyota Motor Corporations ad-tech operations fee.
Advised on agency wide DMP and IT/ taxonomy governance for people analytics to drive quantitative business decisions (i.e.: kNN regression model in Python to predict resource needs for new marketing campaigns, reducing scope by 20%).
Modeled and tuned XGBoost marketing mix model of marketing channels (i.e.: broadcast/cable TV, Print, Social, radio, OOH, etc.) to predict annual vehicle media spend.

Managed the planning, strategies, agendas, and roadmaps for +10 ongoing 20th Century Fox media projects simultaneously.
Reduced PO costs 25% across +1M SKUs (+90k titles) by automating workflows in R and building custom forecasting methods using Holt-Winters Seasonal exponential smoothing.
Participated & collaborated across various departments to ensure successful mapping, development, and execution of promotional sales plans, and regularly pulled and reported on supply chain and promotion / sales performance using SQL.

Worked on numerous projects for digital SVOD service in collaboration with Production, Business Affairs, Content Acquisition, and Marketing senior leadership.
Used R and SQL in AWS RDS to analyze production data in support of new developments for operations improvement.
Owned the design, development, and maintenance of ongoing metrics, analysis, queries, Tableau dashboards, etc.

Implemented data mining methods in R to interpret product association and market research efforts for Stax Records music catalog.
Extracted first-party social data to design customer profiles and report the customer journey from online activity.
Researched Markov chains multi-touch attribution to identify and report marketing channel conversion and effectiveness.

The MLADS Conference is a community-driven event featuring dozens of technical talks, tutorials, and lightning talks on topics such as AI & Data for Good; AIOps; Customer Analytics; Data Engineering & Quality; Deep Learning; Forecasting; Intelligent Edge; NLP; Optimization; Recommender Systems; Reinforcement Learning; Responsible AI; Security Analytics; Speech, Image, & Video Processing; Visualization and more. The Kusto Big Data Analysis engine is widely adopted inside and outside Microsoft due to its rich and intuitive query language and amazing speed of processing complex queries over huge data sets. This tutorial showed how data scientists and researchers can leverage Kusto for their day-to-day research and iterative algorithm development cycle by explaining Kusto native ML and time series analysis capabilities (including anomaly detection, forecasting, root cause analysis, sequence mining and more), show how it integrates with Python/R/Spark and other tools and technologies, and present real use cases that were built on top of Kusto.

Presentation at the Data Science Salon conference in Los Angeles, CA

Analysis of Disney's top grossing films (adjusted for inflation) in Python, using regression to attribute film genre to success. The project includes using regression on the data, as well as bootstrap regression to determine confidence intervals of the intercept and coefficients.

Created multi touch attribution models for marketing campaigns, including Markov chains, using Python. Includes heuristic (ie: first, last, and linear touch attribution) and stochastic (ie; Markov chains attribution) models for attributing customer conversions to marketing channels.

My first go at a Python XGBoost model, using Amazon SageMaker, EC2 instances and S3 buckets. Used to prepare, partition, train, tune, predict and evaluate model. Project involves predicting customers who sign up for a financial product at a bank.

Used NLP techniques (tokenization, stemming, vectorization for TF-IDF) and clustering algorithms (Kmeans and Hierarchical clustering) in Python to mine the "similarities" between films based on their plots provided by IMBD and Wikipedia. The dataset contains the titles of the top 100 movies on IMDb.

For Blacks in Tech Columbus (Ohio) and Women in Analytics

Used the apriori algorithm in R to leverage the associative rule mining to predict online consumer products that are likely to come with the purchase of another item. Identified item associations along with their support, confidence and lift metrics. Created performance plots and interactive associative rule app using plotly and aruleViz packages.

Performed sentiment analysis and LDA NLP techniques to model the script of "The Matrix" using R.

Leveraged SQL and Python to analyze web-scraped data on historical National Football League (NFL) Super Bowl games, performances and broadcasting networks.

Presentation at SatRDay (UCLA R Language Conference)

Capstone Submission #2 for the Harvard University Professional Certificate in Data Science Used R to explore large data set from China’s largest independent big data service platform, TalkingData, to predict which advertising clicks are fraudulent. Used statistical modeling, feature engineering, and machine learning techniques such as decision trees, random forests, linear SVM and kernal SVM. Achieved an accuracy of 96% with an F1 score of 98%.

Capstone Submission #1 for the Harvard University Professional Certificate in Data Science. This capstone is project 1 out of 2 for the final course in the Harvard University Professional Certificate in Data Science. The goal of the capstone projects are to utilize skills gained in the program, including data visualization, probability, inference and modeling, data wrangling, data organization, regression, and machine learning. For this project, I created a movie recommendation system using the MovieLens dataset in R. The version of movielens included in the dslabs package (which was used for some of the exercises in PH125.8x: Data Science: Machine Learning) is just a small subset of a much larger dataset with millions of ratings. I will use the 10M version of the MovieLens dataset to make the computation a little easier. Inspired by the popular Netflix Kaggle competition that challenged users to create an improved recommendation system (that improved the streaming company's then algorithm's error rate by at least 10%), this assignment aims to accomplish a similar task. The optimal goal is to create a recommendation system algorithm for the provided users that would effectively recommend movies based on the provided features, and to optimize such system with an RMSE score <= 0.87750. I achieved an RMSE of 0.876.