The Study.
The study is a peek into my mind. What I’m reading, the research questions I keep returning to, and notes from the academic side of my life.
Bookshelf.
Pick a spine to see why it earned a place here.
6 finished
Finished, 2026
The Wealth of Nations
Adam Smith
As expected, Capitalism from first principles is a pretty great read.
Open questions.
Problems I keep returning to, that I believe will push the frontier of human technology.
Active
Is there a fundamental limit to the generalization capability of Transformer based Large Language Models?
We've seen massive gains in domains like coding and mathematics over the past few years form Transformer based LLMs, but its unclear whether such gains can be made in other domains. It will be necessary for us to consider whether there is a fundamental limit to the LLM architecture, or whether any domain can be solved with the right training data.
Exploring
What exactly causes LLMs to develop an impersonal and distinctly robotic style, and how can we avoid it?
Producing aligned LLMs that still sound authentic and unique is a core issue for building any communication platform that's centered around AI. I'm deeply interested in exploring how we can build authentically human models that can still embody the current formal LLM tone when necessary.
Exploring
What is the optimal memory paradigm for Agents that need to intimately understand their users' lives?
When building any memory system, the core problems come down to understanding which memories are actually important and useful when interacting with the user, and which ones can be forgotten. This is extremely difficult to determine in practice, yet highly necessary for any true personal assistant.
Academic Endeavors.
Research
Computer vision research with Prof. Noah Snavely
Cornell University, Aug 2023 to present
Improving the accuracy of a video entity motion-detection pipeline: implementing a CoTracker optical-flow stage for data preprocessing, optimizing network efficiency, and investigating how the pipeline’s bijective mappings to a 3D canonical space can be used for generative video editing.
Education
B.S. in Computer Science, minor in Applied Mathematics
Cornell University
Merrill Presidential Scholar Award Recipient for outstanding academic and research achievement