blog-admin.zixuanzheng.com Open in urlscan Pro
3.145.61.239  Public Scan

Form analysis
0 forms found in the DOM

Text Content

You need to enable JavaScript to run this app.
Zixuan Zheng
About
Experiences
Projects


ZIXUAN ZHENG

COMPUTER SCIENCE + STATISTICS & MACHINE LEARNING
CARNEGIE MELLON UNIVERSITY

Machine Learning Enthusiast🔮, Web Developer💻, Golf Athlete🏌️‍♀️

ABOUT
EXPERIENCE
PROJECTS


ABOUT

Since high school, I have been actively strengthening my software engineering
skills and working to become a better global citizen. My first full stack
project of LatinGuru: Motto offers me a vision where my creativity can prevail
and contribute to real world impacts, and I was deeply intrigued.

As a junior at Carnegie Mellon University, I continue to become a better version
of myself, actively involved in projects and research, ranging from web
application to natural language processing to robotics.

In my free time, I am also a national-level golf athlete ⛳️ and a movie lover
🎬.

EXPERIENCE
May 2023-Present
Research Assistant · Robotics Institute at CMU
Summer Undergraduate Research Apprenticeship

Engineer a novel approach to incorporate physics constraints with computer
vision algorithm NeRF (Neural Radiance Fields) to accurately predict the state
of cloth-like objects under manipulation in biomedical applications

Design and implement script using Blender Python API to automatically generate
120+ simulation scenes in Blender for algorithm evaluation; incorporated
randomized cloth size, texture, falling heights, and ground objects

Combine Real-Time High-Resolution Background Matting model with COLMAP pipeline
using Python to generate synthetic image dataset with subtracted background and
JSON file of camera Field of Views (FOV) and relative poses

Leverage AprilTag to recover objects' world frame translation and scale to
improve robots' perception capabilities

PythonNeRFBlenderCOLMAPAprilTag
June 2019-March 2021
Research Assistant · Guangzhou Intelligence Electrical Technology Company, Ltd.

Used LabelImg to manually label and classify 1271 sample bird nest images taken
by drones as training and testing dataset

Augmented data through mirroring, rotation, Gaussian blur, pixel removal using
PyTorch, resized images, and labeled regions

Evaluated model's performance using Mean Average Precision (MAP), successfully
raising 10+% accuracy rate and boosting 80 % processing efficiency

Published 10+ pages report in IEEE Access Journal (Volume: 7); granted National
Patent for Technological Invention

PythonData ProcessingLabelImgPyTorch
View Full Résumé
PROJECTS
Portfolio Nexus: Web Portfolio and Administration Hub

Developed and deployed a well-functional single page web portfolio application
using React (TypeScript) and an admin site using Ant Design with integrated
Markdown editor to manage dynamic web page data

Incorporated responsive design, CSS animation, dynamic theme, and a rolling
bullet curtain feature for skills in portfolio page

Designed a well-structured database schema utilizing Python-based ORM library
SQLAlchemy to ensure efficient data storage and management

Implemented the backend functionality using FastAPI framework to enable smooth
data retrieval, manipulation, and secure authentication; utilized Axios for
seamless communication with the backend API

PythonSQLAlchemyFastAPIResponsive Web DesignReactTypescript
COMPASS (COntext Marking and starter PhrAses for Synchronized Socializing)

Collaborated in the implementation of a web-based interface designed to
facilitate real-time online communication for users with speech-generating
disabilities

Performed text cleaning and stop words removal on 3940 dialogue scripts from
switchboard dialogue act corpus using Python

Refined keyword extraction model KeyBERT and evaluated its performance on the
switchboard dataset using cosine word embedding similarity; achieved 78% average
accuracy in extracting conversation topics

Developed visualization pipelines in R for analyzing user satisfaction,
interface usability, and feature effectiveness based on the results of the user
study on 6 pairs

PythonKeyBERTRData Visualization
LatinGuru: Motto - Online Interactive Latin Motto APP

Web-crawled 572 + school mottos and Latin-word dictionaries using Python and
stored data in MySQL

Built a mobile application with Django as the backend to support data
categorization and transmission; manually edited English and Chinese
translations and composed grammar references on the Django admin site

Designed and implemented UI that supports Motto Search, Latin English/Mandarin
Dictionary, Favorites, and Grammar using Vue.js to elevate Latin's accessibility
for beginners and promoted interactive Latin learning

Launched the App on Google Play and received Authorship of Software Issued by
the National Patent Bureau

PythonMySQLDjangoVueReact
Dynamic Memory Allocator

Constructed a 64-bit struct-based dynamic memory allocator that supports malloc,
calloc, realloc, and free functions in C

Boosted space efficiency by reducing internal and external fragmentations
through footer elimination, mini-block implementation, and better-fit searching
algorithm

Optimized throughput from 100 Kops/s to 16,000 Kops/s using the segregated free
list and LIFO insertion policy

C