About Me
My name is Alexander Poon, and I am a rising high school junior at the Edison Academy Magnet School, where I major in Electrical Engineering and Computer Science. Outside of class, I am a student researcher passionate about applying artificial intelligence and machine learning to real-world challenges—right now, I am focused on addressing food insecurity through supply chain optimization. I would say three areas that really define who I am are computer science, research, and community service. I’d love to share more—would you like to hear about a specific area before we go into introducing my projects?
Alexander combines technical skill, research curiosity, and community-driven impact across his initiatives. In computer science, he participated in NASA Space Apps by developing a tool to compare ground-based observations with Landsat data, earned third place in FBLA’s Website Coding and Development event, and sharpened his algorithmic thinking through the American Computer Science League (ACSL). His research experience spans multiple programs, including NonTrivial, the New York Academy of Sciences, and the Science Mentorship Institute, where he explored topics from ethical AI to cognitive science. In the community, he co-lead Revvifi—building websites for over 10 small businesses and coordinating 100+ volunteers—and co-founded FoodFlow, a food redistribution initiative that has donated more than 100,000 pounds of food to local shelters by partnering with businesses to reduce waste and support families in need.
See more in his resume below:
Skills
Technical Skills
- HTML
- CSS
- JavaScript
- React
- Node.js
Design Skills
- UI/UX Design
- Responsive Design
- Adobe Photoshop
- Adobe Illustrator
- Figma
Other Skills
- Project Management
- Communication
- Problem Solving
- Teamwork
- Leadership
Projects
HACKJPS SUBMISSION
CollegeDebtSolver is a web application built to help students explore the potential of AI-powered stock predictions as a means of funding their education—without relying on burdensome student loans. Developed using SvelteKit, TailwindCSS, DaisyUI, and FastAPI, the app pulls financial data from sources like Polygon and Yahoo Finance to forecast whether capital gains over a one-month period will be positive or negative. While limited access to financial APIs and insufficient computing resources posed challenges to fully training the AI model, the team successfully built a polished frontend and functional backend infrastructure. We learned the fundamentals of TensorFlow and developed a roadmap for enhancing the app.
View Project
CONGRESSIONAL APP
Therapute is an AI-powered mental health app developed for the Congressional App Challenge to address the growing inaccessibility of traditional therapy. The app integrates three key features: an AI chatbot that offers personalized emotional support, a video journaling system for users to reflect on their thoughts, and an analytics engine that tracks emotional trends over time. Built with design tools like Figma and Jamboard, the app prioritizes accessibility by allowing users to engage via text or video. Its underlying AI model leverages natural language processing and emotional sentiment analysis to deliver a human-like, therapeutic experience aimed at making mental health support more personalized, affordable, and scalable.
View Project
NORTH JERSEY STEM FAIR
Building the Self-Awareness of Large Language Models under Lossy Compression explores how compressing large language models (LLMs) affects their ability to maintain coherent and accurate outputs. By training three models—LLaMA, GPT-2, and Mistral—using 4-bit quantization and 50–75% weight pruning, the project simulates real-world constraints like limited hardware or deployment on edge devices. Soft prompts were applied to each model, and their performance was evaluated through perplexity, accuracy, and coherence scores, visualized in a confusion matrix. The research provides insight into how lossy compression may introduce cognitive dissonance in LLMs, ultimately informing future strategies for building.
View Project