As a lead machine learning engineer on Einstein Copilot for Tableau, I enhanced the RAG backend of Tableau's natural language interface and vastly improved Copilot's ability to respond to more natural, ambiguous, and underspecified user queries, while also developing instrumentation and analysis tools for the team to measure the impact of their features.

As a lead engineer on Tableau's prototyping team, I identified areas of opportunity for ambitious experimental prototypes, and designed and implemented proofs of concept. I primarily lead efforts in applied ML and NLP, across spaces such as search, knowledge graphs, and human-in-the-loop topic modeling.

I served as a key server engineer on a wide range of features for Facebook Groups, including formatted posts, poll integrity, a new backend framework for the members list that enabled easier ranking experimentation, notification delivery optimization, optimization of the placement of engagement units in the groups feed (top stories, events near you, etc.), and internal tooling for visualizing and analyzing the effectiveness and reach of our sentiment campaign (i.e. a set of video units designed to improve sentiment and engagement with groups).

Developed tooling for quickly onboarding retailers to an experimental platform for serving recommendations from their product catalogs, personalized for each viewer.Contributed to a redesign of the ML models for the platform to be more privacy-sensitive.

Researched operator pushdown optimizations and introduced new optimizations to Tableau's federated query engine, improving the performance of aggregation queries by up to 90%.Supported a large effort to integrate the highly performant Hyper database into the Tableau product.

Contributed to the development and optimization of Tableau's federated query execution engine.

Developed backend support for the analytics pane, a feature for streamlining the process of adding analytical features (reference lines, trend lines, etc.) to a visualization.Developed "instant analytics" features which allow the user to visualize reference lines and trend lines for a selected subset of their data quickly and easily.Contributed to a large effort to burn down long-standing technical debt and improve the quality of the visualization code base.