Kris Selberg
Founding Member of Technical Staff @ UniversalAGI
Hi! I'm Kris. Currently, I'm researching and building AI agents that actually work. I speak okay Swedish, good English, and terrible Spanish. Please reach out if you want to chat about anything!
What am I doing now?
Last updated: July 6, 2025
What have I done? (Me in 10 Seconds)
Flew to the Middle East for a month to build an on-prem AI solution running on real government data
Mar 2025Joined a stealth AI startup as the 1st engineer building AI agents for F500 companies + governments
Dec 2024Spent a semester in SF working on a stealth AI project and flew to Qatar to pitch it
Oct 2024Worked with Sebastian Thrun and Eric Schmidt on applications of video generation models
May-Aug 2024Published a paper in AR + flew to Hawaii to present it at CHI
May 2024Backpacked Western Europe with my best friend for a month
Jul-Aug 2023Ran a marathon with two weeks notice
Jan 2023Lived in Colombia for 3 months coding and creating content
Jun-Aug 2022Grew a TikTok profile to 50k+ followers in 21 days talking about nonfiction books
Aug 2021Interned at the Qatar Computing Research Institute researching and building deepfake detection models
May-Jul 2021Won an applied math competition with 3000+ students
Mar 2020Read a book a week for a year (mostly self-help and "wantrepreneur" books)
2020Made $1k+ selling fidget spinners from China to people in my school
2017Click here for Me in 10 Minutes.
Pet Projects
Breaking the Plane
Breaking the Plane is an augmented reality (AR) application designed for the Meta Quest 3 headset using Unity, focusing on enhancing the comprehension of multidimensional mathematical concepts. It integrates a "Wizard of Oz" optical character recognition (OCR) solution for scanning handwritten input, enabling users to visualize 3D mathematical functions in an immersive AR environment. The application uniquely combines OCR with 3D AR visualization, a feature not previously explored in educational AR systems in mathematics. Its effectiveness was evaluated through a within-subjects study, where it demonstrated significant advantages in engagement and similar ease of use compared to traditional methods like Geogebra 3D (desktop equation visualizer) and Geogebra 3D Calculator (mobile AR).
LLM Princeton Student Search
LLM Princeton Student Search is a web application that simplifies the process of searching and managing information about Princeton University students. It combines NextJS and React for the frontend, OpenSearch for handling search queries, and utilizes AWS services for cloud hosting and deployment. The app features the integration of the GPT-4 API to convert natural language queries into structured search formats, enabling efficient and accurate data retrieval.
The Pi Face Jukebox
The Pi Face Jukebox is a Raspberry Pi application I developed for pranking some friends. This project blends a facial recognition SVM model with customized mp3 playback, playing specific theme songs for individuals as they are recognized by the system. The application uses Python, OpenCV for facial recognition, and the Pygame library for audio handling, along with custom scripts for dynamic music selection based on facial recognition.
M3 Challenge Winner
In March 2020, I worked in a team of five to solve problems related to the electric trucking industry, including calculating optimal placements of charging stations on certain routes, predicting the population of electric trucks 5, 10, and 20 years from now, and determining which routes are most important for developing charging stations first. We ended up placing first in the nation out of 760 teams and winning $20,000 in scholarship money. Our paper can be seen here published in SIAM Undergraduate Research Online, and a video of our presentation can be found here.