AI Art Advisor

To illustrate a potential application of AI models to benefit artists, I created a prototype "AI art advisor" using the latest vison-language model GPT-4o ("omni") from OpenAI. This new GPT model now has vision capabilities, so, I was inspired to ask the question: how well can it understand art?

I created a simple user interface where the user can upload any image (png or jpg format) of their art, along with their API key and any specific requests. In the backend, these inputs are parsed into several API calls to GPT-4o. In the API call, the system is told its role as an art advisor that gives constructive feedback (both negative and positive). Finally, the response is returned and formatted into clean markdown format before being printed out in the Gradio app interface.

Virtual Classroom

This is a virtual reality (VR) project created in the Unity game engine throughout the Spring 2024 semester for the course CMSC838C: Advances in XR. I led a group of 5 classmates to design a virtual classroom environment, and we conducted a user study to investigate the benefits of XR education over traditional remote learning mediums such as video.

Improving Control and Consistency of Diffusion-Generated Video

Semester-long course project created for CMSC720: Foundations of Deep Learning. Investigated state of the art diffusion-based video generation methods, and experimented with various approaches to improving their quality in terms of consistency and controllability. Wrote and submitted a NEURIPS format paper for a mock-conference within our class.

Augmented Reality for Enhancing Elderly Interaction with Touch Screens (Idea)

Hypothetical research project for CMSC839A: Embodied Media Design.