Portfolio

Introduction

I still remember Gamal Abdel Nasser's rise and fall as vividly as a film - not because I'm a history fan, but because of how I learned it. In fall 2021 at LSMSA, Dr. Kyle Stephens taught the modern Middle East as a line of vivid stories where historical figures were drawn out as very real people who had desires and fears, where locations became stages for their defining acts. It was much more similar to watching a play unfold in front of me than any lecture I had seen before. That class taught me what 13 years of Louisiana public school couldn't: learners can rise to any challenge if they enjoy the process. Once I saw what learning could be, I couldn't unsee what was missing everywhere else. With my CS background, I knew I could create something that could scale these experiences beyond elite magnet schools and into anyone's hands. I started with Aluminotes, using LLMs to transform hastily-typed notes into structured lecture materials - born from my friends' and my ADHD making traditional note-taking nearly impossible. We used it for a class project, and even won first place at a pitch competition with it, so we were very happy with the results. I joined my college's ML research laboratory to learn more about AI, as I knew I would be using it later on to personalize education. My research focused on curriculum learning and experimental training methods for convolutional models - work that led to presenting at AICS at IBM's Thomas J. Watson Laboratory. Now, as fullstack engineer and researcher at BeNakama, I work on Ascend, a math-based RPG that adapts difficulty to student skill - directly implementing a program that tests students at their abilities while keeping them engaged. This is also the most direct application of my research - I am involved with design and implementation of the systems that the students use. I'm pursuing research to understand the cognitive and technological foundations of effective learning, and find out how to apply that to allow everyone access to it. My goal is to make Dr. Stephens' classroom the norm, not the exception - so every student can experience what it feels like to genuinely love learning.

Aluminotes

AccessibilityEducational TechnologyCompetition Winner

The Aluminotes home page, with the logo on the right (a black paper crane against a bright red-orange gradient) and information about the website on the left.

During my sophomore year of college, my friends and I with ADHD were struggling to keep up with traditional note-taking. Our notes were broken and unreadable after lectures, making review nearly impossible. I realized that given the topic and the notes as context, an LLM (which were at the time still extremely experimental) could transform it into actual structured notes, making review more accessible for students with attention disorders.

We built Aluminotes on Django, using Google Gemini to process and reformat the notes. The system would take in shorthand notes and output coherent lecture summaries with proper structure and context.

We used it for a class project and refined it based on our own experiences as the target users. Being in our own user target allowed us to focus on the specific pain points we encountered daily.

Aluminotes won first place at our school’s Inn-eaux-vate pitch competition. The judges recognized both the technical execution and the genuine accessibility need it addressed. More importantly, it proved to me that EdTech could directly solve real barriers to learning when designed by those who experience those barriers.

You can find the code at https://github.com/DudeSquaredEdud/Aluminotes

BeNakama

Backend DeveloperEdTechUK Schools

A screenshot of the Ascend combat system, with the protagonist (a knight) on the left, an enemy (a bush monster) on the right, and the attack prompt at the bottom, with the question "Which of these numbers is a multiple of 2?"

After my undergraduate research on curriculum learning, I wanted to apply those principles at scale. I sought a position where I could implement adaptive learning systems that balance challenge and engagement - the core insight I gained from Dr. Stephens’ class. BeNakama’s mission to prioritize genuine learning over compliance metrics aligned perfectly with my philosophy.

As fullstack engineer, I work across the entire stack to build Ascend’s combat and adaptive difficulty system. I lead backend design remotely, architecting systems that track student performance across mathematical concepts and adjust problem difficulty in real-time.

The RPG format keeps students engaged while the underlying algorithms ensure they’re working at their optimal challenge level. I’m involved in both the design decisions and implementation, allowing me to ensure the technical architecture serves the pedagogical goals.

Ascend represents my first professional implementation of research-backed adaptive learning. The system is designed to identify individual student strengths and weaknesses, then use that data to create personalized learning pathways. My goal is to prove that technology can scale the kind of personalized, challenging, engaging education I experienced at LSMSA to students in UK high schools and beyond.

You can sign up at https://play.benakama.com

Wavelet Analysis Research

Machine LearningUndergraduate ResearchIBM Presentation

The logo for the Deep Learning and Data Analysis Lab at The University of Louisiana, Lafayette. The logo itself is a popout effect from a wavy, purple and blue background, with the letters "D" "A" "D" "L".

Undergraduate research on optimal training order for image classification neural networks, using wavelet entropy as a proxy for the difficulty for a neural network to successfully classify a given image. Presented these findings at the IBM Thomas J. Watson Laboratory.

Although I had help out in the DADL lab before, this was the first project where I was more in the know about the process than the other researchers. Wavelet complexity was brought up in a meeting after thinking about my favorite GIMP image effect, “wavelet decomposition”.

My mentor at the time, Gabriel Trahan, had experience in signal theory. He knew that wavelets were used as a means of extracting signal from noise, but had never thought about them in an image context. I knew about them from the image point of view, but never knew how they worked.

In short, wavelet decomposition is a more advanced edge detection algorithm that matches along directions and scales. These directions and scales are “taken out” of the image, for the process to be repeated. This allows you to capture details about the image that are not visible when all the scales are presented together.

We had decided to use wavelet entropy, a special method of calculating the difference between scales, as the proxy for image complexity. Initial testing showed promising results, but an oversight - unnoticed groupings - had resulted in the model performance to be worse overall.

I am still proud of the experiment and I even have the poster on my wall. I believe I learned a valuable lesson on how simple considerations are often lost when thinking about the big picture.