I am currently pursuing a Master's degree in Software Management at Carnegie Mellon University. I received my bachelor's degree from Sun Yat-sen University.

I am skilled in software engineering and system design in CPP and Python. In particular, I used to design and implement a low-code AI platform with a unified algorithm toolbox for in-house engineers, leading to a great saving of engineering time. Currently, I am interested in distributed system in deep learning to enable larger foundation model to be trained and deployed efficiently.

   /      /   zexiny@andrew.cmu.edu

• Designed a mechanism to train a deep learning model with a few lines of configurations.
• Designed a unified code structure to extend modules efficiently.
• Designed an industry oriented usage design to ensure every model is stabled and reproducible.
• High-quality implementation of data augmentation algorithm with CUDA to enable training faster.

• Generality and superiority of the platform were shown by achieving SOTA on accuracy and computational efficiency for various benchmark.
• Designed a novel semantic segmentation network structure that can accurately identify tiny defects, a critical bottleneck for achieving 99.99% yield index.

• Reproduced SOTA algorithms in unsupervised anomaly detection.
• High-quality implement of algorithm to enable training in larger image with less memory and training a model in less than 2 minutes.

• Designed an efficient AutoML framework to support any tasks with many parameters to tune and enable extending black-box optimization algorithm efficiently.
• Designed and implemented hyper-parameters optimization technology and achieved the same or better performance without the involvement of AI engineers in more than 20 in-house real-world datasets.

• Introduced and optimized neural architecture search technology and achieved 50% less onboard inference time (crucial for industrial applications) while preserving the same performance.

• High-quality implementation of common CPU and CUDA ops for computer vision algorithm.
• An easy-to-use library for in-house engineers to deploy an efficient SDK for deep learning model inference in Python.

• Designed and implemented an AI model manager platform that integrates with the existing infrastructure and codebase.
• With this platform, AI engineers can monitor, manage, visualize and track models they have run without adding any lines of code.

• AI engineers can customize the platform with a few lines of code to deploy their specific algorithm solutions.
• Clients can efficiently manage large-scale datasets and create AI models with the benefit of AutoML technology, thus developing their own data-driven AI models.