Stanford University
Kundaje Lab
Dr. Anshul Kundaje
March 2024 - Present

Integrating genomic knowledge graphs into a clinician-accessible LLM chatbot for precise genetic variant discovery and rare disease diagnostics

Stanford University
Dynamode
Dr. Emily Fox
January 2024 - March 2024

Analyzed short-term trends of Continuous Glucose Monitoring (CGM) data to predict long-term Hemoglobin A1C test results using machine learning time-series modeling

Stanford University
Dynamode
Dr. Emily Fox
September 2023 - January 2024

Used LLMs to analyze diabetes patient’s CGM graph data and generate clinician-style routine messages on behalf of Certified Diabetes Educators (CDEs) to summarize patient trends and make recommendations

Georgia Institute of Technology
Robot Autonomy and Interactive Learning (RAIL) Lab
Dr. Sonia Chernova
August 2021 - May 2023

Using activity recognition and activity summarization to detect anomalies in an elderly patient’s daily routine by implementing DBSCAN in Python, as well as NLP to summarize the relevant part of the patient's day to notify their remote caretaker. With the increase in elderly patients in independent living, this project helps the patient maintain a daily routine and alert the caretaker of any anomalies, as it is imperative for patients with MCIs or Dementia to maintain a daily routine for their long-term well-being.

Georgia Institute of Technology
Polo Club of Data Science
Dr. Duen Horng (Polo) Chau
January 2022 - May 2023

Created an in-browser Visualizer for Neural Network Training by using Uniform Manifold Approximation and Projection (UMAP), a dimension reduction technique, to visualize how an object detection model perceives and changes inputs throughout different layers of the neural network by using Python and JavaScript.

Paper Publication - VIS 2022

Georgia Institute of Technology
Polo Club of Data Science
Dr. Duen Horng (Polo) Chau
October 2020 - May 2023

Defending against adversarial attacks for Computer Vision-based object detection models in collaboration with Intel. Created a web-based tool that uses POSE Estimation to detect key joints on a person’s body using Python doing a specific action to see how the key joints are perturbed due to an adversarial Patch or PGD attack to inform researchers who work in the adversarial attack space about key areas of attack by common Patch and PGD attacks.

Live Demo
GitHub Repository

Georgia Institute of Technology
Polo Club of Data Science
Dr. Duen Horng (Polo) Chau
August 2020 - May 2023

Using mixed reality to better prepare surgeons so they go into the surgery knowing what cuts need to be made. Working with the Children’s Healthcare of Atlanta (CHOA), we created an iOS mixed reality application using Swift and ARKit where surgeons can upload a custom 3D model of their patients’ hearts and perform practice surgery.

Paper Publication - Asian CHI 21
Paper Publication - VIS 2022

Georgia Institute of Technology
Polo Club of Data Science
Dr. Duen Horng (Polo) Chau
August 2020 - December 2020

Built a website that uses AI to automatically crop photos into headshots by detecting the keypoints on a person's face and aligning them to photography's rule of thirds. Within seconds and without ever sending images to a third-party server or the cloud, Magic Crop creates the perfect headshot for social media and directories using HTML and JavaScript. Over 500 active users including multiple colleges at Georgia Tech

Live Website
News Article

Stanford University
Stanford Medical and Stanford Machine Learning Group
Dr. Andrew Ng & Dr. Matthew Lungren
July 2019 - August 2019

Using the dataset of pictures of pre-implantation embryos, pre-trained an encoder with a convolutional neural network (CNN) based architecture, PredNet, to repeatedly predict what the embryo would look like in five minutes until the model could determine whether the embryo would lead to a successful pregnancy.

Stanford University
Stanford Medical and Stanford Machine Learning Group
Dr. Andrew Ng & Dr. Matthew Lungren
July 2019 - August 2019

Classified knee abnormalities using a convolutional neural network based architecture called PredNet by extracting features from axial, coronal, and sagittal knee MRIs of 100+ patients.

Harvard University
Harvard Medical and Spaulding Rehabilitation Center
Dr. Randy Trumbower
January 2018 - May 2020

Created a novel Augmented Reality (AR) walking physical therapy solution using the Microsoft HoloLens. Using the Microsoft HoloLens, with Unity for the front-end design and C# for the back-end algorithms, we created an immersive, engaging, and fun AR walking physical therapy solutions for patients, especially children, where the user feels like they are in a game, but in reality they are strengthening their legs and re-learning how to walk.

Paper Publication - Frontiers in Human Neuroscience