HI, I'm Elizabeth Orrico! I work at the intersection of AI research, engineering, and education. 

Before working in industry, as a Machine Learning Engineer at KLA, I taught computer science at the University of Virginia and high-school mathematics and physics through Teach For America. My current work focuses on generative AI, agentic systems, semantic analysis, and interpretable machine learning. 

I'm particularly interested in agentic systems, interpretability, evaluation, representation, and how increasingly capable models change the way people interact with information. I most enjoy working on ambiguous problems where the answer isn't already known: I care about both what AI systems can do, and how we can understand what they are doing.

I created the Semantic Universe Explorer (SUE) to solve a problem I was encountering: how to meaningfully compare semi-structured documents such as corporate sustainability reports. In doing so, it became a natural showcase for several of my core AI specialties, including vector embeddings, retrieval-augmented generation (RAG), and semantic exploration. 

After winning the annual Hackathon at KLA for creating a project centered around AI-assisted Global Impact Reporting, I became very interested outside of work in how I could scrape these reports to gauge different corporations' performances in metrics relating to subjects such as: 

• Water Usage

• Hazardous Waste Handling

• Workforce Management and Fair Labor

and many more.

This ever-evolving library of viewers and tools demonstrates some of the ways to impartially compare not just passages from Global Impact Reports but any document type where important quantitative information must be parsed from passages written in qualitative, natural language.

My work with multi-agent systems goes back to 2015, when I earned my Master’s degree in Computer Science at the University of Southern California, specializing in these formulations.

Supported by a fellowship from the Department of Homeland Security, I studied a special game-theory formulation known as a Stackelberg game, in which two competing agents, an attacker and a defender, act and react to one another.

My research applied this framework to TSA airport screening, where defenders must allocate limited screening resources despite uncertainty about when passengers will arrive. I designed algorithms intended to preserve strong defender performance despite imperfect information.

I left the project in 2016. The research continued and was published shortly after my departure.

View a job talk and learn more by clicking here.

When it comes to multi-agent systems, I definitely have favorites: I enjoy systems that thoughtfully employ adversariality, asymmetric information, and agentic collusion.

I incorporated many of these techniques while designing this prototype system, which can readily be used as a general purpose anomaly detector, or binary classifier. 

This tool is especially useful in 2 cases: (1) when semantic information from the data needs to be taken into account-- such as a fraud detector that has built-in awareness that a purchase in the middle of the night might be suspicious, and (2) when interpretability and audit trails are necessary for the classification.

I've always really enjoyed a great data visualization, whether that is for education, featured in the news media, or for technical, research, or workplace consumption.

Over the years, I've gotten to create some insanely fun ones myself. Spanning from the microscopic-- my award-winning senior thesis that visually tracked the movement of dendritic cells in atheroscleretic patients, a collaborataion with La Jolla Institute-- to the cosmic-- applying 3D algorithms to astronomical blueshift data for the National Radio Astronomy Observatory (NRAO).

I wanted to share a few here, as well as any other miscellaneous work.

I designed these materials for undergraduate computer science students studying discrete mathematics, including logic, proof techniques, induction, counting, probability, relations, and graph theory.

The curriculum emphasizes the transition from intuitive mathematical thinking to formal reasoning and proof, with examples built around the places students are most likely to get stuck. The materials include lecture slides, exercises, examples, and other resources developed for university computer science courses.

I created and filmed Generative AI Is for Everyone!, a four-part, corporate training series designed to make generative AI understandable and useful to employees across KLA.

The approximately 2.5-hour series began with Journey to GenAI, an accessible introduction to how generative AI works and why it matters. It continued with Prompting Like a Pro, followed by more advanced instruction on enterprise AI tools, agents, and vibe coding.

More than 1,000 employees watched the first installment, and Generative AI Is for Everyone! became KLA’s most popular training offering that year.