CV

University of Southern California

Master of Science in Computer Science

• Affective Computing
• Analysis of Algorithms
• 3D Graphics and Rendering
• Multimedia Systems Design
• Autonomous Cyber-Physical Systems
• Innovation for Defense Applications
• Advanced Mobile Devices and Game Consoles
• Information Retrieval and Web Search Engines

National Yang Ming Chiao Tung University

Bachelor of Science in Computer Science

• Parallel Programming
• Computer System Administration
• Introduction to Machine Learning
• Introduction to Database Systems
• Introduction to Operating Systems
• Computer Vision for UAV Autopilot
• Advanced Programming in the UNIX Environment
• Data Structures and Object-oriented Programming

Advantech

Software Development Intern

• Developed a configurable PyTorch computer vision toolkit for classification, detection, and segmentation
• Modernized ResNet training for CUDA 11 on Ampere GPUs, supporting 5 variants and variable-sized images
• Built MobileNetV3 Faster R-CNN and DeepLabV3 workflows with YAML configs, 5 optimizers, and 3 schedulers
• Converted LabelMe to COCO and deployed a Detectron2 RetinaNet solder-joint model to Inspection AI

MugArchive

• Built a React app to catalog Starbucks Discovery Series mugs with an interactive US map and location markers
• Integrated Supabase PostgreSQL with realtime sync and a local fallback dataset
• Added a collection dashboard with total count, category breakdown, and quick status visibility
• Delivered dual grid and map views with instant search and category filters for states, cities, and special locations
• Automated GitHub Pages deployments with GitHub Actions, rebuilding only upon relevant source changes

ParkMarks

• Built a React app to explore and track 60+ U.S. national parks with an interactive map and clickable markers
• Integrated Supabase PostgreSQL data access for dynamic park records with a local fallback dataset
• Implemented park detail cards with images, visited status, visit dates, ratings, search filters, and progress tracking
• Added shareable deep links, browser history support, and responsive dark mode with localStorage persistence
• Shipped updates to GitHub Pages using an artifact-based GitHub Actions pipeline with concurrency control

DreamMesh: Blender Plugin for AI-Driven 3D Scene Generation

• Built a Blender 3.0+ add-on in a four-person team to generate multi-object 3D scenes from text prompts
• Used OpenAI API to convert user prompts into structured JSON scene specifications
• Automated a 3-step text-to-3D pipeline using Hugging Face, Stable Fast 3D, and Blender asset import
• Developed Blender UI panels and operators to manage the generation workflow and object placement

Wi-Fi Rate Control on Ryu and OVS

• Built a system-level Wi-Fi rate control service on Linux using Ryu, Open vSwitch, OpenFlow, and TCP/IP
• Automated collection of 4 telemetry fields including MAC, IP, RSSI, and SNR for controller-side decisions
• Parsed metrics in the Ryu controller to apply SNR-based rate and MCS selection, enforcing QoS and flow rules
• Reduced controller overhead by batching OpenFlow rule updates for throughput and latency optimization

Ransomware Propagation and Payload

• Automated SSH dictionary-based login attempts via shell scripts to simulate a ransomware attack
• Overrode the default "cat" command to enable worm-like propagation across devices
• Encrypted JPEG files via a deployed payload and triggered a custom ransom pop-up message
• Coordinated a two-person team using Git for version control and collaboration

Accelerating K-means Clustering with Parallel Implementation

• Architected three parallel K-means algorithms for distributed execution using OpenMP, MPI, and a hybrid model
• Wrote shell scripts to automate dataset generation, runtime logging, and parallel execution across configurations
• Benchmarked large-scale datasets of 20 million points with MPI, achieving up to 7.4x speedup
• Collaborated in a three-person Git-based team to evaluate scalability and profile performance bottlenecks

Analysis of EEG using Deep Learning

• Coordinated with a team to design an EEG experiment using E-Prime to collect signals during index finger lifts
• Preprocessed EEG data with EEGLAB, including filtering, artifact rejection, and channel segmentation
• Developed and trained an EEGNet classifier in PyTorch to predict motor intent prior to finger movement, achieving approximately 72.5% accuracy through cross-validation and confusion matrix analysis

When BERT meets Google Comments

• Built a BERT-based NLP classifier to predict sentiment and category labels from Google Maps reviews
• Developed modular data collection, preprocessing, training, and inference workflows in Python
• Scraped and structured review data with BeautifulSoup for supervised model training
• Achieved approximately 90% test accuracy using Google star ratings as weak supervision labels
• Integrated trained model inference into a PyQt5 GUI for real-time user-facing predictions