Winona Chu | 朱紜彤

yuntungc@uw.edu yuntungc winona1111 Medium

About Me

I'm Winona from Taiwan, currently pursuing my Master's in Electrical and Computer Engineering at the University of Washington.
I'm a Graduate Research Assistant in Professor Jenq-Neng Hwang's Information Processing Lab, specializing in computer vision and object detection. I work with YOLOv8 models, synthetic data generation pipelines, and generative AI for 3D environments. My technical toolkit includes Python, Java, PyTorch, TensorFlow, Docker, AWS, and React.
Previously, I interned at Innodisk optimizing YOLO models and at Hualiteq building IVR backend systems. I also won 1st Place in the 2023 National Universities Innovation Competition for developing a color vision assistance MR app.
During my free time, I enjoy playing basketball, watching TV series, and taking city walks. I'm lowkey trying to record life in US on my website (there's not much there yet though!). I also write about some of my experiences at UW on Medium!

M.S. / ECE / @UW-Seattle

Skills

Education

Experience

Category: All Industry Research Course Project

May 2026 - Present

Khorium

Remote, US

Software Engineer Intern

Engineered full-stack fix for a 1.6× overestimation in mesh cost prediction: replaced bounding-box surface area proxy with OCC-derived CAD face areas in Python backend, piped through PostgreSQL JSONB to a Zustand-managed React/TypeScript frontend, and recalibrated runtime model from empirical production data.

Unblocked cross-repo deployment failures in a multi-service AWS CDK pipeline by diagnosing a broken Alembic migration and resolving a cascading generated-type conflict.

Full-stack Python React TypeScript PostgreSQL AWS CDK

Jul. 2025 - Present

Information Processing Lab, University of Washington

Seattle, WA, USA

Graduate Research Assistant

[CVPR 2026 Workshop] First-authored research on zero-shot single-temporal anomaly detection in remote sensing using CLIP + DINOv3 dual-vision fusion (BSSRS), achieving 95.38% Pixel AUC.

Fine-tuned a lightweight Qwen3.5-VL-4B with LoRA on MixedWM38 (38k+ images) for semiconductor defect VQA, optimizing the model footprint for high-throughput inference; applied GRPO with custom reward functions to mitigate hallucinations and resolve multi-label combinatorial explosion.

Architected scalable distributed training and inference pipelines on HPC clusters including Taipei-1 (NVIDIA DGX SuperPOD) and NCHC; integrated Flash-Attention 2 and vLLM to maximize PagedAttention and VRAM utilization for high-throughput multimodal workflows.

Engineered automated synthetic data generation pipeline producing 50,000+ images, integrating Vega Prime rendering and GroundingDINO (Open-Set Detection) for auto-labeling, reducing manual annotation time by 50%.

Optimized dataset composition to achieve 84% mAP on YOLOv8-tiny across five environmental conditions, meeting constraints for edge deployment.

Deployed SAM2 (Segment Anything Model 2) for zero-shot semantic segmentation of microscopic particles and engineered a geometric filtering algorithm using OpenCV to decouple boundaries from cracks.

Independently developed novel integration of Generative AI Foundation Models (Hunyuan3D-2) with depth-based layering in Blender to synthesize photorealistic 3D environments for training data.

Computer Vision Object Detection YOLOv8 Synthetic Data Generation SAM2 Generative AI CVPR LLM VLM HPC

Sep. 2025 - Dec. 2025

Distributed Systems: Sharded Key-Value Store

Seattle, WA, USA

Course Project

Architected a fault-tolerant, linearizable key-value storage system using Multi-Instance Paxos for consensus and log replication, ensuring data consistency across server failures and network partitions.

Implemented a fault-tolerant Shard Master service to manage dynamic re-sharding and load balancing, enabling replica groups to join/leave the cluster without service downtime.

Engineered cross-shard ACID transactions using Two-Phase Commit (2PC) with locking to ensure serializability and deadlock prevention in a distributed environment.

Distributed Systems Paxos Java Consensus Protocols

Jan. 2025 - Jun. 2025

UW Capstone Project: A Cloud-Based Multimodal Data Ingestion Platform for Time Series, Cycling, and Open-Circuit Battery Data

Seattle, WA, USA

Full-stack Engineer

Built a multimodal data ingestion platform with React and FastAPI, exposing 15+ RESTful endpoints to ingest, standardize, and query battery datasets across 6 file formats, backed by AWS RDS (PostgreSQL) and S3.

Developed LLM-powered client-server data standardization pipeline using OpenAI API for semantic schema alignment, increasing the column matching rate to 90%.

Architected a modular data processing pipeline (extract → clean → standardize → split) triggered via FastAPI endpoints and backed by S3-based file storage and PostgreSQL metadata tracking.

Implemented the frontend using React with dynamic routing and secure authentication via OAuth2, providing clients with isolated access to project-specific data.

Designed and implemented comprehensive RESTful API architecture with 15+ endpoints handling multimodal data ingestion, user authentication, and real-time processing status tracking.

Deployed and configured AWS EC2 instances with PostgreSQL database integration, ensuring secure cloud infrastructure for production-grade data storage and retrieval.

Software Development Data Science Web Development REST API

Apr. 2022 - Dec. 2023

NCCU Graduate Project: MR Application - Color vision deficiency Assistance

Taipei, Taiwan

Software Developer

Developed a color adjustment software module for Hololens2 mixed reality goggles, designed to assist individuals with color deficiency in specialized environments such as chemical laboratories, utilizing Unity and C# technologies.

Implemented TensorFlow Object Detection API to accurately identify laboratory glassware, achieving a precision of 94% mAP in object recognition.

Built a partial frame color adjustment function using a custom color transformation matrix within the Unity development environment with C#.

Optimized device integration through Socket programming and C# Coroutines, successfully reducing screen synchronization latency between remote devices and local servers by 100 ms.

Secured First Place in the 2023 National Universities Innovation Competition.

GitHub Repo => UnitywithOD

Software Development C# Unity TensorFlow Edge Device

Aug. 2023 - Dec. 2023

Hualiteq International Co.

Taipei, Taiwan

Software Research and Development Intern

Developed backend logic for an IVR system using JavaScript and MSSQL, and implemented dynamic voice response generation through Google TTS API integration, enabling real-time, personalized user interactions.

Documented system integration workflows and GitLab CI practices for team-wide adoption.

Software Development JavaScript MS SQL IVR Technical Documentation

Jul. 2022 - Aug. 2022

Innodisk Co.

Taipei, Taiwan

Software Research and Development Intern

Built a real-time screw welding anomaly detection pipeline targeting edge deployment, expanding the labeled dataset from 1.3k to 7.6k training images (~5.6x) via geometric and noise-based augmentation.

Fine-tuned YOLOv4-tiny using a Keras-based pipeline deployed to Xilinx KV260 via Vitis-AI (.h5 → q.pb → .xmodel), achieving 100+ FPS at 8W on DPU with 90% mAP@0.45 post-quantization.

Developed Python-based unit testing for the team project, enhancing error detection.

Containerized project using Docker on Linux to reduce environment-related issues during deployment.

Machine Learning Object Detection Python Unit Test Linux

Apr. 2022 - Jun. 2022

Transnational vaccination verification system

Taipei, Taiwan

Back-end Developer

Developed a comprehensive web platform providing integrated travel and vaccination information for cross-border entry requirements during the COVID-19 pandemic.

Implemented a robust MySQL database to centralize and manage international travel and entry regulations, utilizing PHP to dynamically render and access critical information.

Designed an interactive user interface featuring registration and commenting functionalities, enabling collaborative knowledge sharing and real-time information updates for travelers.

Software Development MySQL PHP HTML

Feb. 2021 - Jun. 2021

Money Grabber

Taipei, Taiwan

Developer

Developed a location-based restaurant discovery application for campus environments using Java in Eclipse, designed to assist new students in quickly exploring local dining options.

Implemented web scraping techniques utilizing Python's Selenium and BeautifulSoup libraries to aggregate comprehensive restaurant information from Google Map.

Enhanced user experience by integrating Google Maps API, providing interactive geographical visualization and refined interface design for seamless restaurant exploration.

Software Development Web scraping Python Selenium API