< Home < Projects

This section gives specific examples of non-research projects I've worked on (to see the projects made specifically for research, go here. This page might include projects I've worked on for classes, but there are several I've done only for fun (but only the best/most representative of these are shown).

Arch Linux Personal Computer

Status: in progress (when is this ever not in progress?)

Timeline: January 2024 to current

There's an adage that "Linux is free if you don't value your time", and unfortunately, this is even more so with Arch Linux. The amount of work this operating system requires would suprise most people, and you might take some things for granted unless you've worked in an environment with much less support compared to Windows (e.g., NVIDIA pushed an update one time that caused Arch systems to stall when their lids are closed). Although, working in Arch Linux is much more enjoyable from a programming and customization perspective; the "everything is a file" design means linking disjoint processes and devices together is very simple, and the open-source nature of the operating system means I can (in principle) control everything that goes on in my computer. It's fun!

Computer Security: Image-based Randomware through Buffer Overflows and Steganography

Status: complete (report)

Timeline: August 2023 to December 2023

Professor: Dr. Kirill Morozov

This was a fun project where we had to create a ransomware (in a virtual environment) that encrypted files with a hidden private key and then decrypt those files at a later date. I wanted to do something weirder than a standard privilege escalation or something, so I made my own image-viewing application with a buffer overflow vulnerability in it. It would read images as "ppm" files, and would use an unbounded "strcpy" operation that would create and execute an arbitrary executable file if the attacker hid binary code within the pixels of an image. It's based off of a real attack done against the Korean version of a 2003 Windows operating system! Alongside this, I also wanted to manually implement RSA encryption, which involved creating my own prime-number generator in C without any external libraries.

Empirical Analysis: Study of Life-like Cellular Automata

Status: complete (report)

Timeline: August 2023 to December 2023

Professor: Dr. Sayed Shah

This project only required we use empirical analysis techniques on some dataset. I was (and still am!) really interested in cellular automata at the time, so I used the "state of the grid at each timestep" as a synthetic dataset and performed various statistical analysis methods to show how different life-like cellular automata variants might change their grid in different ways. I liked all the graphs I made! I secretly want to do more projects involving statistical analysis so that I can create and interpret new graphs.

Algorithms: Review of Classical Verification of Quantum Computations in Linear Time

Status: complete (report)

Timeline: August 2023 to December 2023

Professor: Dr. Yuan Li

I had an ulterior motive while completing this project: we were meant to do a thorough review of an article from FOCS, but I wanted to additionally learn more about quantum computing. I chose one of the only quantum papers at this conference, which is not a very good introduction to the topic! I read a few other books at the same time to help piece things together--including Quantum Physics: What Everyone Needs to Know by Michael G. Raymer (whom I attended a seminar of in 2023!)--and this presentation became both an introduction of quantum physics to the layman audience and a paper review like it was supposed to be in the first place.

Deep Learning: Creating a Programming Language from Neural Turing Machines

Status: complete (report)

Timeline: January 2023 to May 2023

Professor: Dr. Heng Fan

This project wanted us to implement a deep-learning system on a new dataset or task. Around the same time I was interested in programming language design and the concept of domain-specific languages (especially esolangs), and wondered if it was possible to create a programming language that would be capable of learning algorithms built from individual discrete instructions, where a human supplies the target inputs/outputs to the algorithm and the model learns which instructions to activate in which order to make that algorithm. The algorithm-learner turned out to be more complicated of a task than I could do in a semester (and it's something transformers have already accomplished, in either case), so I pivoted to making a programming language instead: the user passes in a list of instructions as plaintext, and the model executes each by passing each through a list of sub-models (with each trained to implement an individual instruction). The langauge works, but is very limited in types of programs it's able to accomplish.

Parallel Processing: Polyhedral Compilation

Status: complete (report)

Timeline: January 2023 to May 2023

Professor: Dr. Song Fu

The polyhedral approach to compilation is a technique where a loop in a program is transformed into an equivalent representation that reduces the value dependencies between loop iterations. It visualizes array accesses as a function dependent on an affine iterator value, and uses the polyhedral method to skew, translate, or shift the affine polyhedra to make the dependencies a different shape with respect to iterator. This project creates my own polyhedral optimizer with the help of different polyhedra-specific libraries like OpenScop, Clan, and CLooG. Although, I didn't realize at the time that this takes a good background in math to understand how to do it! Specifically, loops require the calculation of a special matrix through a linear-equation solver, which is difficult to compute without knowing what the "Farkas lemma" was (which I didn't...). My solution was to randomize this matrix, compiling the resulting program, and running a collection of tests to ensure the final program was still semantically correct. This means the compiler has the possibility of generating programs that don't match what the programmer intended (so it should not be used in the real-world), but it does generate programs that are significantly faster in certain cases!

Undergraduate Thesis: Syntax as a Tool of Thought

Status: complete (report)

Timeline: January 2022 to December 2022

Advisor: Dr. Barrett Bryant

This was my final requirement for completing my Bachelor's degree. I started it with Dr. Barrett Bryant in his class on compilers as a step in graduating with Honor's, and it led into an entire thesis in the following semester. It's composed of two parts: first, an analysis on programming languages and their ability to solve domain-specific problems; and second, the introduction of a new programming language created by me which demonstrates flexibility in syntactical design as to accommodate different kinds of domain-specific problems. The thesis is submitted alongside a compiler (lexical, syntactical, and semantic analysis) and an interpreter for the language, both written in standard Python. The language is designed around allowing the user to create their own syntactic constructs, with things like for-loops and if-statements not being defined in the language by default. The user can create functions that utilize a bytecode to implement these important constructs and any more the programmer might need. The end result of this is that very diverse problems can be elegantly represented in the language in the format they naturally relate to; I demonstrate this by creating a program that looks nearly identical to SQL, which the programmer implemented as a result of defining a significant portion of the SQL syntax from within the new language.