Statement on Generative AI
I realize that today, with the advent of “vibe-coding” it is easier than ever to develop small scale projects for a portfolio. With that in mind, I want to make it clear that, for my recreational work, I make it a point to use as little generative AI as possible throughout the development process.
I do these projects for fun, and for the learning experience. In the rare cases where I do use generative AI, I use it as a last resort, and I make a real effort to learn from it rather than just prompting it to complete the project – in my eyes, that defeats the purpose.
For the following projects, I always aimed to minimize the use of generative AI while prioritizing the learning process over the “finished product”.
Finetuned BERT classifier that determines what school of philosophy a piece of text is closest aligned with

- Curated dataset containing text entries that represent four schools of philosophy
- Finetuned DistilBERT model on dataset
- Achieved F1 scores >= 0.99
- Hosted on Hugging Face Spaces with Gradio UI
Web puzzle game where you string together two-word phrases to navigate from one word to another

- Generating small dataset of two-word phrases
- Constructing graph object containing word relationships and navigating with DFS to obtain chain of words
- Implementing front-end interface with Hyperdiv framework in Python
Master’s Thesis, Institute for Artificial Intelligence, University of Georgia

- Preprocessing EEG data from DEAP dataset
- Extracting continuous wavelet transform features from EEG data using fCWT library
- Training vision transformers from vit-pytorch on CWT features
- Finetuning pretrained vision transformer for private dataset
Personal project

- Implementation of NEAT (Neuroevolution of Augmenting Topologies) algorithm
- Evolving neural network topologies based on success in retro video game environment
- Evolved network to play retro games like Ms. Pacman
Personal project

- Creating Snake from scratch using C++ and SFML library
- Implenting AI-mode with greedy pathfinding algorithm
Evolutionary Computation Term Project

- Adapting Blondie24 algorithm to a new environment
- Programmed a neural network model to evaluate board states during minimax search
- Trained network weights using evolutionary strategies
- Compared Blondie24 algorithm to modern evolutionary programming algorithm