I am a student passionate about learning and making change to this world. My work primarily focuses on various aspects of cybersecurity and machine learning, but I love to learn about and do many other things as well.
June 2022 - Present
Portland, Oregon
The Institute for Computing in Research is a consortium of researchers, students, and teachers who work on pedagogical aspects of using advanced computing methods for research and scholarship. Any application of computing to physical science, life science, social science, arts and humanities is of interest to our researchers.
June 2024 - Present
June 2022 - August 2022
May 2022 - August 2023
Mount Sinai, New York (Remote)
The BioMedical Engineering and Imaging Institute (BMEII). BMEII focuses on the use of multimodality imaging for brain, heart, and cancer research, along with research in nanomedicine for precision imaging and drug delivery. BMEII is composed of research groups in all aspects of imaging research.
May 2022 - August 2023
September 2018 - June 2023
Portland, Or
GoodTime Chinese School (GTCS) is the largest private Chinese-English bilingual school located in Portland, Oregon.
September 2018 - June 2023
|
|
2024-2027
B.S. Computer Science & B.S. MathematicsCGPA: 3.9 out of 4Taken Courses:
Extracurricular Activities:
|
|
|
|
2020-2024
High School DiplomaWeighted GPA: 4.5 out of 4Taken Courses:
Extracurricular Activities:
|
Stanford Computer Vision - Best Pose-Estimation Project. This is an AI dance coach that extracts poses using
1) MoveNet-based key point extraction
2) A custom Transformer model to distinguish key dance move figures
3) Automated error identification using Dynamic Time Warping
4) Use templated errors and LLMS to create user friendly feedback
5) Annotated video interface for feedback
An interactive AR-based learning platform designed to guide and educate car owners through essential maintenance tasks, such as oil changes, brake replacements, and more. Controlled using speech commands and hand gestures, and contains interactive step-by-step guidance with video overlay assistance and step contextualization.
Originally made a robot to do handwriting for me, it can also reproduce artworks in cool ways as well.
Built a website to features Westview Bridge Club blog posts
This website was made as a creative response to reading Beloved by Toni Morrison. It is also a part of a school project.
A Debian linux systems challenge involving standing up and security a Squid Proxy Server, LEMP stack, system corruption recover, and malware forensics.
Funbridge is an online bridge game platform that provides 2 free deals every 24 hours. Thus, we built a web automation script to automatically collect the daily free deals. The script has been working for 2+ years and has helped 20+ accounts collect deals.
A Windows Server 2022 collection of cybersecurity challenges involving ransomware analysis, digital file forensics and recovery, keylogging, REGEX, and memory forensics. With the goal standing up and securing an OpenSSH, IIS webmail server and MySQL database on the server.
My side project for the Institute for Computing in Research. Without hacking the quizlet match game, I built a bot using tesseract OCR, jaccard similarity tests, and computer inputs to gain superhuman speeds in the Quizlet Match game.
A Ubuntu systems challenge collection including forensics and SSH challenges.
A LEGO robot that can play chess against use and uses computer vision to detect position of the game. It can be connected to lichess.org to play chess online.
EV3 mindstorm LEGO Rubik’s Cube Solver inspired by https://nerya21.github.io/MindCuber
Worked with NASA and LANL scientists to model spurious solar wind forces measured by the Advanced Composition Explorer satellite onto the Laser Interferometer Space Antenna Pathfinder satellite. These models are used to predict space weather effects on the Laser Interferometer Space Antenna for possible use as a serendipitous space-weather observatory.
Deep learning in medical imaging often requires large-scale, high-quality data or initiation with suitably pre-trained weights. However, medical datasets are limited by data availability, domain-specific knowledge, and privacy concerns, and the creation of large and diverse radiologic databases like RadImageNet is highly resource-intensive. Our solution, RadImageGAN is the first multi-modal radiologic data generator trained on the RadImageNet dataset, capable of producing high-resolution synthetic images across 12 anatomical regions, 130 pathologies, and 3 imaging modalities. When combined with BigDatasetGAN, it enables automatic pixel-wise annotation for segmentation tasks, significantly boosting model performance while reducing the need for manual labeling.
We leveraged DenseNet121 (a deep learning model) with separable convolutions, to identify anatomical landmarks and measure leg length from frontal CT scout radiographs, optimizing for minimal mean absolute error (MAE) through tuning of learning rate, batch size, and patch size. Results demonstrate that separable convolutions outperform global average pooling in both accuracy and convergence, offering an efficient, automated solution for assisting radiologists in diagnosing leg length discrepancies.
We developed an automated radiomics analysis system that combines U-Net-based thyroid segmentation with radiomics feature extraction to classify thyroid nodules on ultrasound images. Compared to a convolutional neural network baseline, our method achieved higher segmentation accuracy (dice scores of 0.77 and 0.74) and substantially improved sensitivity, negative predictive value, and positive predictive value, while reducing the false negative rate by 41.1%, highlighting its potential as a noninvasive diagnostic tool.
