Over the years I have done a lot of really cool projects. This is as close as I can get to a comprehensive list of the things which I am able to share publically on the internet.
I researched methods of classifying material phases from atomic coordinates using neural networks at the University of Rochester Laboratory for Laser Energetics, supervised by Shuai Zhang.
Replicated results from prior work on classifying material phases using Dynamic Graph Convolutional Neural Networks (DGCNNs).
Developed a modified version of DGCNN which can obtain a better test accuracy.
Invented Gaussian Kernel (GK) architecture, which obtains a significant performance improvement with comparable test accuracy to modified DGCNN.
Researched scaling of GK models with respect to various hyperparameters.
I created a search engine focused on Wikipedia pages. Primarily written in Python, this search engine creates embeddings for text and images using open source CLIP models, which can be searched using a vector database. The various components of the search engine are compartmentalized as Docker containers. Using containers, managing the complex parts of the search engine can be made easy. For more information on this project, please view the project's public codebase.
I researched methods of managing applications using containers to support scientific computing projects at the University of Rochester Laboratory for Laser Energetics, supervised by Richard Kidder. Research is presented in my project report.
Naturify is an iOS app that can be used for identifying images of plants, animals, and fungi. The app utilizes a neural network (visual transformer architecture) that I trained in order to perform image classification.
This is an offline version of Naturify.
Image Garden is an iOS app that allows you to perform semantic searches on the images in your photo library. It uses a neural network (CLIP) to perform the semantic search.
Earch Explore is an iOS app that classifies birds based on their calls, using a neural network.
WordLab is an app that allows users to take notes, and then as AI questions about their notes.
Trained neural network to predict nearby species given latitude and longitude coordinates.
Trained neural network to perform image classification on a dataset of 10000 different species (plants, animals, fungi etc.) for my app Naturify. I used a visual transformer architecture, and trained the model with Pytorch on an Nvidia A100 GPU for several days.
The code for this is not currently available to the public.
Created backend server for Naturify. This functions as an inference server for the iOS app, and hosts the app's website. This code is not currently available to the public.
Floral is a Jax based python neural network library. Using Jax as a backend, it allows you to create, train, and save neural networks. Floral can be intstalled via Pypi.
this C program will encode any file into a DNA sequence. The idea is that some day when data can easily be written to DNA, something like this can be used to encode files into a DNA sequence so that they may be stored in living organisms.
My attempt at creating an operating system. Code written in C and assembly creates a basic bootloader, and screen driver.
Motivated by understanding how financial exchanges work, I created a currency, and security exchange for items in my apartment. This includes a working stock exchange which uses and order book which can be used to buy and sell securities with the currency.
Several experiments that I did on neural networks for my own curiosity.
Created a basic web crawler in Rust as a prerequisite to creating my search engine.
Creating this website took a lot of effort. You can check out the repo for the code here.