Luke Zheng

Machine Learning Engineer Intern @ vivo

About

I am currently pursuing a Bachelor of Science in Computer Science at UC San Diego, where I enjoy working at the intersection of software engineering and applied AI. I take pride in learning quickly, experimenting with new tools, and bringing technical concepts to life through real applications. My industry experience includes developing video generation and 3D scene understanding pipelines as a Software Engineer Intern at Vivo. I transformed multi-modal visual data into reliable 3D point clouds, benchmarked alternative representations for more accurate scene modeling, and helped standardize evaluation using Fréchet Video Distance. These contributions strengthened the team’s ability to train and compare advanced generative models at scale. At UCSD’s Graduate and Undergraduate Economics Lab, I led development of a Chrome automation tool that simplifies job applications by autofilling forms across 15+ platforms. I worked directly with users to iterate on key features, improve reliability, and publish clear documentation that made adoption seamless. Across my roles, I have built strong communication skills by collaborating with researchers, engineers, and non-technical stakeholders. I enjoy turning ambiguous problems into structured workflows and measurable progress. I am eager to join a team that values curiosity, thoughtful engineering, and delivering products that make a tangible difference.

Country

United States

City

San Diego

Industry

Computer Software

Skill

Point Cloud Library (PCL), 3D Reconstruction, Transformers, AIGC, Recommender Systems, Machine Learning, PyTorch, Logistic Regression, C++, Algorithms, Coding Experience, Programming, Object-Oriented Programming (OOP), JavaScript, XML, Data Structures, Oral Communication, Software Development, Python (Programming Language), Software Construction

Experience

Machine Learning Engineer Intern

vivo

2024-9 - Present · 2 yrs 1 mo

• Engineered a 3D scene reconstruction pipeline that integrated visual-inertial SLAM, depth sensing, and camera intrinsics to generate accurate point cloud representations from iPhone Pro data. • Enabled structured language understanding of indoor environments by deploying Meta’s SceneScript model on egocentric video and depth data • Automated preprocessing of egocentric video streams, aligning multi-sensor data for consistent transformer-based layout predictions. • Validated model outputs through MeshLab point cloud visualizations, identifying edge cases and refining scene layout consistency. • Wrote technical documentation on pipeline architecture and inference flow to support model reproducibility and future deployment in AR/VR research.

Software Engineer Intern

vivo

2024-6 - 2024-9 · 4 mos

• Designed and engineered a deep learning pipeline using Python, PyTorch, and open-source models to train a Vector Quantized Generative Adversarial Network (VQGAN) on the UCF-101 action recognition dataset, enabling a video-generation AIGC model capable of synthesizing human action clips from latent representations • Integrated MAMBA, a lightweight sequence modeling module, into the VQGAN inference workflow to enable faster real-time inference and significantly reduce model latency in a production-grade video generation system • Automated the preprocessing of video frames by building Python-based scripts that streamlined data cleaning, annotation parsing, and frame extraction, reducing total preprocessing time by approximately 30%

Machine Learning Intern

Institute of Intelligent Computer Technology

2023-7 - 2023-9 · 3 mos

Suzhou, Jiangsu, China

Programmed a YOLOX machine learning model using PyTorch for highly accurate labeling of components for fault detection Processed 200+ images with LabelImg to create a learning dataset be used to train an AI model Coded a function to process txt and xml files into different formats using Python