Theo Urban

Learner / Leader / Runner

Carnegie Mellon University AI Major

Director of ScottyLabs' Labrador Committee

15-281 Teaching Assistant

Hello! I'm Theo Urban, a student at Carnegie Mellon University studying Artificial Intelligence. I'm passionate about the power of the internet to solve problems, and endlessly intrigued by CS theory, math, and society.

About

I was born and raised in Pittsburgh, PA (Go Steelers!!), and I'm currently a Junior at Carnegie Mellon University studying Artificial Intelligence. My interests lie in the intersection of real-world problems with interesting algorithms. At CMU, I've delved into the cutting edge of Machine Learning at the same time as I pursue the fundamentals via my discovered passions of CS Theory and Math.

I've found great fulfillment in teaching as a TA for CMU's classical AI class (15-281) and developing projects that I, and others find useful. ScottyLabs, CMU's largest software engineering club, provided an incredible space for me and a team to develop CMUMaps, a project I love for the complexity of it's theory and extreme utility. To help others have a similar experience to my CMUMaps experience, I now lead ScottyLabs' Labrador committee, an incubator-style program for the future of ScottyLabs (and external) projects.

Beyond my coursework and clubs, you can find me running all around Pittsburgh and in local races. I love running for the dedication it requires, the exhilaration of success, and the peace of the outdoors.

Lazarus AI SWE Intern

At Lazarus AI, I contributed to the company's document understanding platform by building a B2B sub-organization management system using Firebase and Flask. This structure allowed for efficient data, usage, and metrics separation and simplified billing management for clients. I also implemented obfuscation techniques for our Retrieval Augmented Generation (RAG) API for on-premise deployment and APAC Cloud deployment, including seamless CI/CD integration. Additionally, I increased test coverage on the RAG API to 80% and others through extensive unit testing, database mocking. I also prototyped a SQL logging scheme, and established stable backups to support the Retrieval Augmented Generation (RAG) microservice. This role enabled me to strengthen my skills in backend development, database management, and production-level deployment for a rapidly evolving AI startup.

CTAT Research Intern

As a Research Intern for Cognitive Tutor Authoring Tools Lab (CTAT Lab) at Carnegie Mellon University, I worked on an innovative tool that empowers educators to create adaptive online coaching systems. My contributions focused on enhancing functionality and user experience by developing a hybrid undo-redo system, intuitive spreadsheet and modal interfaces, and window-swapping capabilities. I also collaborated on refactoring 3,000 lines of code from JavaScript to TypeScript, which improved code quality and reliability. Additionally, I resolved 51 bug reports, strengthening the application's performance and stability. This role offered valuable experience in full-stack development and the opportunity to contribute to a meaningful educational technology project.

CMUMaps

Indoor-outdoor mapping and navigation for the CMU campus, as well as room availability, events search. I started working on this in Fall 2023 in ScottyLabs, and then recruited and lead an incredible team in Spring 2024. Released in Summer 2024 with 2000 users!!

StravaAnalysis

Prediction and Analysis of my Strava running data -- The background of this website is a model of me running around Pittsburgh. It is a great way to test techniques I learn, including route generation via next-token-prediction and NLP run type classification.

WikiFinding

Scraped and created a graph of a subset of Wikipedia, which gave me the idea to try to beat the wikipedia game (ie. node-to-node pathfinding) using embeddings of the pages, not via content but just graph context (neighbors, neighbors of neighbors...) I had some success but intend to pick this back up soon.

And More...

My other projects include a Running EsoLang, starter code for ScottyLabs' Labrador Committee, a ScottyLabs Superapp, this website, and probably something new by the time you're reading this... check out my GitHub at github.com/tsurbs