I’m Zhiyu, a Biomedical Engineering grad at Columbia with a passion for building tech at the intersection of health and computing. I work across brain-computer interfaces, AR/VR, and medical imaging, using tools like Python, Unity, and C# to create practical, human-centered solutions.
Designed to streamline financial analysis and reporting, this system automates data processing, AI-driven insights, and Power BI visualization for business decision support.
Built to support cognitive neuroscience experiments requiring memory recall assessment and EEG recording, this task challenges users to memorize object patterns and locations, then recall them from memory.
Migrated the task from Python to C# to run in Unity VR, enabling cognitive experiments integrating both EEG and eye-tracking data collection with a VR headset.
Developed to demonstrate real-time animal image classification using a TensorFlow CNN model and a user-friendly interface.
Designed to streamline medical image preprocessing and augmentation, featuring an interactive interface.
Curated and maintained large-scale digital pathology image databases with precise labeling and quality control to support AI model training. Collaborated with AI engineers to refine image analysis and enhancement algorithms, integrating AI-generated outputs with original scans to build comprehensive reference libraries for model validation and research.
Optimized the training efficiency of a CNN–Mamba hybrid model for 3D MRI segmentation by eliminating software bottlenecks and maximizing hardware utilization. Implemented CPU-side data preloading and a custom Dice Score computation to significantly reduce GPU-memory overhead and runtime latency, cutting total training time by 75% while maintaining a Dice Score accuracy of 0.988.
Conducted pre-scan checks, prepared patients for MRI procedures, assisted physicians in precise positioning, and configured the MRI machine’s starting parameters.
Engaged visitors with educational demos on robotic technologies, promoted company products to hospital directors, and established key contacts for future collaboration.
Analyzed feedback and clinical operation data from four major exhibitions to identify performance trends and develop two data-driven enhancement plans adopted by the team. Designed and implemented a hierarchical, parameter-based data-sorting model that improved inventory management efficiency by 50%. Additionally, engaged over 50 visitors daily through data-informed demonstrations of robotic technologies and product promotion, securing 13 key hospital-director contacts for future collaboration.
Courseworks:
Courseworks: