About Me
I'm a UChicago graduate (2024) who studied economics and data science. I am now pursuing an MS Thesis in computer science at Columbia, where I am interested in leveraging information theory & deep learning to better understand modern AI systems and the brain. I hope to do contribute research in the effor to enhance the communication process between humans. My dream would be to enable the transfer of high dimensional information directly between humans, reaching a sort of advanced telepathy.
Featured Projects
Information Loss Dynamics in Iterative VLM/Diffusion Transformations
Project exploring how information is lost vs. preserved when transformed between images and text. We test this by captioning an image with a VLM, compressing the visual representation into a textual one. We then use the caption to condition an image generation through a diffusion model, and caption this new generated image. We compare the semantics between both the images and captions, and repeat this process over many iterations. We find that certain semantic attractor basins form, where image clusters get tighter with each iteration.
Skills: Python, PyTorch, VLMs, Diffusion Models, Information Theory
Mechanistic Interpretability - ResNet
Project exploring how a custom-trained residual CNN learns interpretable features and classifies images from the CIFAR10 dataset. Analyzed feature learning heirarchies and circuits through activation visualization.
Skills: Python, PyTorch, Feature Visualization, CNN, Residual Networks
Facial Tracking Research
A research project with the UChicago EMK Lab using geometric facial tracking meshes to analyze conversational facial interactions
Skills: Python, Data Preprocessing, MediaPipe, PCA
Andrej Karpathy Learning Series
A comprehensive series of projects following Andrej Karpathy's educational content, exploring deep learning fundamentals through hands-on implementation. Includes both the makemore language modeling tutorials and GPT implementation from scratch.
Skills: Python, NLP, Deep Learning, Neural Networks, Transformers, PyTorch, Attention Mechanisms
LLM Political Biases
A project fine-tuning LLMs on FOX/CNN data to explore ease of model biases.
Skills: Python, LLMs, Fine-Tuning
Other Projects
Mechanistic Interpretability - Superposition Hypothesis
Project recreating Anthropic's "Toy Superposition" paper, showcasing how ReLU pins layer activation spaces to the "privileged" model neuron basis.
Skills: Python, PyTorch, Sparse AutoEncoder, Deep Learning
Retrieval Augmented Generation Demo
A project implementing a RAG framework for document information retrieval and synthesis.
Skills: Python, LLMs, RAG, LlamaIndex
Graph Neural Network
An exploratory GNN project analyzing the arxiv research paper dataset.
Skills: Python, PyTorch, PyTorch Geometric, Deep Learning, GNN
Neural Signal Clustering and Gaussian Mixture Models
A project exploring clustering methods on neural signal data such as KMeans and PCA-assisted KMeans. Additionally explores use of same techniques on mnist digits dataset.
Skills: KMeans, PCA, SVD, Neural Data
Image Compression and Dimensionality Reduction
A project exploring Singular Value Decomposition and its use in compressing image file sizes.
Skills: Python, Linear Algebra, SVD