Projects

GoDaddy Airo, Supabase, Hugging Face

P3 Myeloma is a web-based, AI-enhanced shared decision-making resource being developed for adults with relapsed or refractory multiple myeloma. The platform incorporates MyeAI, a conversational AI tool designed to help patients explore treatment-related information, support informed decision-making, and ultimately reduce decisional regret. My work has focused on translating the research and study requirements into the patient-facing digital experience, including consent-first onboarding, participant account creation and authentication, access controls, accessibility and usability, privacy considerations, participant engagement tracking, and integration of the AI tool into the broader website. The project is currently in the development and pre-pilot phase, with ongoing work focused on refining the chatbot experience, connecting cited resources, expanding its knowledge base, and preparing the platform for participant testing.

Python, Hugging Face Transformers, PyTorch, FastAPI, Docker, AWS

PolyText is a multilingual NLP API built with FastAPI that provides sentiment analysis, named-entity recognition, and automatic language identification through a unified interface. Its model registry dynamically routes requests to language-specific models or multilingual fallbacks, while a production-oriented architecture emphasizes modularity, dependency injection, lazy model loading, and extensive automated testing. The project demonstrates how modern NLP models can be engineered into a reliable, extensible ML service rather than isolated model scripts.

Python, C++, NumPy, SciPy, pandas, scikit-learn, Matplotlib

I built a lightweight human activity recognition system that classifies everyday physical activities directly from raw three-axis accelerometer measurements. Rather than relying on the precomputed features provided with the UCI HAR dataset, I developed the signal-processing and feature-engineering pipeline from the raw sensor signals and evaluated models using subject-independent testing. The project also explores the tradeoff between model accuracy and resource usage for embedded systems.

HTML/CSS, JavaScript, Vercel, Git/GitHub

My first personal website was an early exploration of combining software development with visual design to create a space for my academic, research, and professional work. I built the site around a photograph I had taken on a trip to Mexico while visiting family, using its colors and visual atmosphere as inspiration for the website's palettes and overall design.