Working in the Cross Channel Marketing team

• Led a team of three in designing a recommendation engine for email campaigns, using XGBoost for conditional average treatment effects and fine-tuning Chat-GPT-3.5-turbo for product description clustering, achieving a 20% increase in conversion rates.• Designed a full-stack model pipeline by integrating the React front-end with the Python back-end, utilizing Hasura GraphQL for seamless data flow with the PostgreSQL database, while managing pull requests and resolving integration issues to ensure smooth functionality.• Deployed the full-stack system on an AWS EC2 container using Docker Compose for container orchestration and Airflow for workflow automation, enabling one-click execution from the front-end web pages, which was instrumental in client demos and securing new contracts.

• I taught Probabilistic and Statistical Decision-Making for Management, an undergraduate course at the Questrom School of Business

• Designed and automated a model report in plain HTML and CSS - to avoid introducing dependencies in client environments - which includes the intermediary and final summary statistics of the data in clear tables and graphs. Data scientists immediately gave positive feedback in how it facilitates error detection in the model pipeline saving the company millions.• Wrote REST APIs in Julia (HTTP.jl) for model configuration of the recommendation system and queried them in a TypeScript (React) application, making new model features accessible and improving the efficiency of the model pipeline.• Initiated a transition from Julia to Python by writing a module that imports the R-learning Python functions into Julia; enhanced the team Jenkin’s environment by resolving outdated packages and upgrading the Dockerfile base image from CentOS7 to Ubuntu 20.04.• Deployed the above applications in client environments using Linux command line tools and shell scripts.

Making better pairs between callers and call-agents.• Assisted the production team with simulating counterfactual models in the A/B testing by re-weighting the outcome of model A under the condition of model B; accelerating the model search process by considering a larger range of models.• Developed an R-learning package to make better pairs between agents and callers for a large enterprise telephone company; built hyper-parameters into the package that optimizes the treatment effect estimation by incorporating caller specific features.• Mentored new team members in understanding the codebase and clarifying the statistical background of the models; Delivered daily updates to the whole team and effectively conveyed the application of treatment effects on caller-agent pairings.

• I provided research assistance to Professor Davide Pettenuzzo on "Dividend Suspensions and Cash Flows During the Covid-19 Pandemic: A Dynamic Econometric Model." Journal of Econometrics, 2022. - Data collection from Bloomberg and Nasdaq - NLP text cleaning and sentiment analysis on Nasdaq data and 8-K Forms• I worked with Professor Tymon Słoczyński on writing the R package version of implementation for the paper "Interpreting OLS Estimands When Treatment Effects Are Heterogeneous: Smaller Groups Get Larger Weights." The review of economics and statistics, 2022. - Package: https://cran.r-project.org/web/packages/hettreatreg/index.html• I worked with Professor Daniel Tortorice on constructing Macroeconomic models of Inflation Expectation utilizing VAR expectation models, and did analysis on Treasury Inflation Protected Bond mispricings.

• Statistics for Economic Analysis• Statistical Modeling with R

• Applied Econometrics with R - Wrote R code that produces all figures, tables, and solutions for "Introduction to Econometrics." Stock and Watson, 3rd Edition.• Computer Simulation and Risk Assessment - Wrote Python code for MA, EWMA, and VaR models

• Built a Markov-switching trading algorithm leveraging Acadian’s Global Risk Index to dynamically adjust asset allocation, optimizing portfolio returns by differentiating between risky and safe market regimes using state-transition probabilities and risk-adjusted metrics.• Backtested the algorithm with monthly rolling portfolio rebalancing and volatility-adjusted position sizing, achieving a t-score above 3.

I worked for the UNA of TB in two appointments, the first was as a liaison between my school's UN club (which I was the founding President of) and the regional association, in this fashion I sought and placed students in various positions in the organization; and helped coordinate over 30 student volunteers along with chairing committees in the model UN for middle and high school students for two years. The second was contacting and and meeting with political leaders, our Representatives in the House and Senators, in the region to promote the UNA agenda.