zephaniahsdata.comzephaniahsdata.comThis page provides a list of all the projects I've completed since I began maintaining this log, as well as those I am currently working on.
I like to think of this page as a journal. I can look back and reflect on my past projects and consider areas for improvement. I can also keep track of what I am currently working on and remind myself to finish what I started.
Each block below is a link. If the block includes a personal project or code I wrote, it links to the project or code. Otherwise, it links to an outside page with information about the item.
Status: In progress
This repository contains an implementation of the paper 'Gradient Routing: Masking Gradients to Localize Computation in Neural Networks'. The code specifically replicates the MNIST autoencoder experiments described in Section 4.1 of the paper. Some training details differ, but this should give a solid start for building on the experiments in the paper.
Status: Completed December 15, 2024
This project introduces a novel approach to protein classification using steering vectors derived from protein language models. The implementation demonstrates this technique using the ESM-2 protein language model, achieving approximately 90% accuracy in protein localization classification with only 200 training examples per class.
Status: Last updated December 30, 2024
A website I created for the UChicago AI Safety Club, a student organization that I help lead.
Status: Last updated December 17, 2024
A website I created for an upcoming AI alignment conference at the University of Chicago.
Status: Finished (for now) as of December 29, 2023
Toy Models of Superposition is a groundbreaking machine learning research paper published by authors affiliated with Anthropic and Harvard in 2022. As a personal ML project, I replicated some of the most important results from the original paper. The project involved training over a million small neural networks.
Status: Finished September 17, 2023
This project contains code for a fully functional neural network using only stdio.h and stdlib.h in C. Matrix multiplication, derivative calculations, and gradient updates were all done without using any outside libraries.
Status: Finished September 4, 2023
I took this course with a group of other interested students to self-study the potential impact of AI. We read technical papers which we discussed as a group to gain understanding. The syllabus for the course can be found by clicking on this block.
Status: Ready for release on August 19, 2023
zephaniahdev.com is my personal website developed using React and Three.js. It is not only a hub for my profiles and projects - it's also interactive and fun to use. It contains fun animations, an interactive user interface, and customizable settings.
Status: Testing release
A personal project to display the data I have collected about my productivity. You are looking at it right now!
Status: ~1/3 Complete
Rust is a systems programming language I am learning for fun. I have experience in C which makes coding in Rust interesting because there are a number of significant differences between the two languages.
Status: ~3/4 complete
This is a book discussing a new theory for how the neocortex works, and how it relates to the potential impacts of AI. I think the book is interesting, but I find the lack of intellectual humility on the side of the author a bit off-putting and untrustworthy.
Status: Implemented eight algorithms so far
Implementation of various sorting algorithms and graphs of their performance in C and Rust.
Status: Completed August 9, 2023
I took this class during the 2023 summer to self-study AI and neural networks. My lab work and final project have been posted on my GitHub.
Status: Version 1.0.1 (stable release August 7, 2023)
tinyfilter is the computer vision equivalent to Andrej Karpathy's micrograd. It converts images into ASCII art using the principles of CNNs (convolutional neural networks). The project has been open sourced on PyPi (to install run pip install tinyfilter). The code is also available on my GitHub.
Status: Completed for release on July 19, 2023
This is a portfolio website for my dad, featuring all his work from the over 30 years he spent in journalism. The site features an intuitive user interface for searching through a large volume of work.
Status: Completed for release on July 18, 2023
zephysballoons.com is a very simple personal website for my balloon twisting business.
Status: Completed June 13, 2023
Using a linear regression model and publicly available data, my letter predictor predicts the final letter in five-letter words.
Status: Completed June 12, 2023
Using python I created a text user interface for the popular game 2048 and wrote an algorithm that beats the game ~30% of the time. The code is available on my GitHub.