I'm a third-year Integrated Science student at the University of British Columbia integrating data science, statistics, and economics. Passionate about data and uplifting under-represented genders in tech, I volunteer with Women In Data Science (WiDS@UBC) to help nurture a diverse tech community at UBC. Further, I love to dance—I manage a high-level student-led UBC street dance team, Project U&G. Welcome to my (work-in-progress) portfolio!
OvaTech AI is an app that aims to spread awareness of PCOS and help patients get their diagnosis faster — surveys indicate that up to 70% of women with PCOS remain undiagnosed and 40% of healthcare providers lack comprehensive knowledge of the diagnostic criteria for PCOS. I worked with a team of 5 members and 2 project mentors to develop two ML models to predict if a patient has PCOS: a Random Forest model with 96% accuracy that operates on the patient’s symptom data and a Convolutional Neural Network using VGGNet16 that operates on the patient’s ovary ultrasound images. The Figma interace I developed includes other features such as daily symptom tracking and educational resources on PCOS for increased DAU.
memoAIzed is a sustainable image generator that reduces energy consumption up to 99%! Image generation is notoriously unsustainable—it can cost as much as 1 full smartphone charge in energy for a single generation. This memoization-inspired image generator uses the power of AWS to efficiently and scalably store previously generated images, reusing them if a prompt passed to the image generator is semantically-similar to previous prompts. In this project, we used Amazon's DynamoDB and S3 for efficient image and prompt storage, as well as Amazon Titan for image generation and Llama 3.1 for semantic prompt matching. Essentially, we've lowered the cost of image generation down to the cost of text generation, which only costs around 1% of the energy. You can imagine how the application would improve dramatically if the user base scaled and the database of previously seen images grew larger and larger. As a bonus, the latency for memoized images is 5 times faster on average than for non-memoized images!
equiHealth is a full-stack MERN application, supported by Spline for 3D graphics. It's a health app inspired by Duolingo in the sense that the user tries to maintain a daily streak, except instead of language-learning, they must carry out a wellness task. Such tasks include drinking enough water, journaling, or taking a walk. The app documents which category of tasks the user most frequently chooses, then recommends Arc'teryx products for the user based on the user's favourite category—the mission of the youCode 2024 Hackathon was to develop new channels for Arc'teryx to consider for their omnichannel marketing initiatives.
