Created instructions, helper functions, and examples that teams across the data organization use to productionize machine learning models using Zapier’s new tech stack. Identified potential roadblocks and provided teams with a clear path forward to ship models at a time where there was a lot of ambiguity introduced by new tooling and in-progress migrations.Met with leaders across product and engineering to understand how the data organization could better enable engineering squads that were experimenting with LLM-powered products. Spent some time embedding with one team as an experiment and consulted with other product squads to understand how data could be used to enhance their products. This ultimately resulted in a team of 6 being resourced to build user-facing AI/ML products and tooling/infrastructure.Tools: Python, SQL, Databricks, Airflow, Django, LLMs, NLP

Worked with marketing, product, UX, and engineering leadership to create and execute on a roadmap to increase the number of merchants who are using Shopify Logistics for their logistics needs. As part of this effort, the team shipped both stakeholder-facing and user-facing products in areas such as lead attribution and marketing channel optimization, marketing personalization, identification of high intent leads, lifetime value estimation, and automatic detection of products that are ineligible for our logistics services (using text and image data from the product description). Tools: Spark, Python, SQL, GCP, dbt, Oozie/Airflow, Docker, git, NLP

Worked on an app that integrated with the website experimentation platform to extend its capabilities beyond simple AB testing. This work enabled product managers to run multivariate experiments as well as experiments that used contextual information about the user to choose the best treatment. Tools: Python, Flask, AWS, git, multi-armed bandits, contextual bandits

Worked on a variety of ML products that integrated with our website to curate and enhance the information that was shown to users. This ranged from text-based models that enabled us to parse and display human generated text in more useful formats to personalized routing engines that directed users toward pages where they were more likely to submit leads in real time.Advised analysts, engineers, and junior data scientists working on machine learning projects such as churn modeling and extraction of text data from complex pdfs. Worked with other data science teams to refactor their codebase to include more modular functions and high fidelity tracking.Presented results at ~10 internal tech conferences, engineering brown bags, and weekly all company meetings. Placed first in one company-wide hackathon and second in another.Piloted tools that were rolled out across the data org such as MLflow. Helped data scientists and analysts on other teams onboard onto new tools and platforms. Tools: Python, Spark, SQL, MLflow, NLP, mixed models, tree based models, Flask

Owned the onsite ad forecast that our sales reps used to determine how many ad impressions were available to be sold for the next year. Created a forecast that helped to guide the transition from impression based monetization to click and action based monetization. This allowed us to identify areas where the new click and action-optimized ad portfolio was likely to underperform and adjust our strategy before the full rollout.Tools: Spark, R, SQL, Databricks, time series forecasting, regression

[Epidemiology Department]Used statistical models to examine whether systemic inflammation during early childhood is associated with breast cancer risk factors that emerge later in adolescence. Understanding this link can help inform preventive health interventions, improve monitoring of at-risk individuals, and reduce long-term cancer risk.Wrote reusable functions to automate tasks such as running models on different subsets of the data, identifying anomalous observations, and producing figures from model output.[Health Policy & Management Department]Used operations research to understand how efficient different countries are at converting resources that are invested into their healthcare systems into positive healthcare outcomes. This information could be used to identify which countries are making efficient use of resources and understand how their practices could be applied to drive better healthcare outcomes in other countries.Amassed data from multiple surveys on country-level healthcare systems and used that data to quantify the resources being allocated towards healthcare for this research.Tools: R, ggplot, regression, nonparametric statistics, operations research

Analyzed gene expression datasets containing 20,000+ genes each with the goal of identifying key genes that regulate environmental stress responses in plants. Understanding which genes regulate stress responses (such as responses to heat stress or drought) furthers to our ability to engineer crops that are more resilient to extreme weather conditions. Used nonparametric statistics to identify genes that are differentially expressed under various stress conditions and unsupervised clustering to identify families of genes with similar expression patterns (that likely work together). Presented results at 6+ departmental events and external research conferences. Completed a 41 page undergraduate thesis.Tools: R, unsupervised learning, nonparametric statistics, clustering

Evaluated a range of materials to understand which would be best suited for growing stem cells in a 3D environment. Growing stem cells in 3D environments (as opposed on flat plastic plates) gives us a better understanding of how the cells would behave within a living organism. This is crucial for studying and developing functional tissue grafts.Presented a poster on this work to a mixed audience with varying levels of technical knowledge at an internal research symposium.

Used image processing techniques to create 3D models of the HIV-1 virus based on 2D images. These 3D models could be used to identify potential drug targets and help researchers understand how new anti-virals might interact with the virus.Presented a poster on this work to a mixed audience with varying levels of technical knowledge at an internal research symposium.