Jin-Xun Zhang

I built an end-to-end deep learning pipeline to support rapid infection-risk screening for peritoneal dialysis in a real-world setting. From an ML engineering perspective, I enforced reproducibility and reliable generalization by versioning data/preprocessing, preventing leakage with case/patient-level splits, and using a fixed held-out test set as the deployment proxy.For model optimization, I benchmarked multiple CNN backbones under a consistent training protocol (same input spec, augmentations, optimizer) to ensure fair comparisons. I used staged fine-tuning (freeze → gradual unfreeze), learning-rate scheduling (warm-up / OneCycleLR or cosine decay), regularization (weight decay, dropout), and early stopping to stabilize training and reduce overfitting. Hyperparameter tuning followed a coarse-to-fine approach: LR range tests and small pilot runs to narrow LR/batch size, then systematic ablations on weight decay, augmentation strength, unfreeze depth, and patch/cropping strategy, validated with repeat runs and error analysis (confusion matrix + FP/FN review).For deployment, I engineered an on-device inference pipeline including preprocessing, patch-based inference, and aggregation into a single risk score, with intermediate outputs retained for debugging/monitoring. I integrated the model into an iOS app, managed TestFlight releases, and handled signing, bundle IDs, and versioning to ship a maintainable, production-oriented ML system.

As a teaching assistant in the lab, I supported hands-on training and daily operation for key equipment and processes, including an in-house thermal evaporation system and a spectrometer for optical characterization. I helped new members set up deposition runs (basic chamber preparation, parameter checks, and process logging) and guided measurement workflows to ensure consistent, repeatable spectra. I also assisted with MEMS fabrication tasks and demonstrated EWOD (electrowetting-on-dielectric) device operation, covering standard operating steps, common failure modes, and practical troubleshooting to keep experiments running smoothly and safely.

I worked as a Process Engineer in the laser drilling area, optimizing machine parameters for PCB panel drilling through structured DOE and data analysis to improve yield, stability, and process consistency. I standardized validated settings into production-ready specifications to reduce lot-to-lot variation and strengthen manufacturing performance. I also supported the solder mask (green paint) department by improving line processes and investigating recurring issues with cross-functional teams (manufacturing, equipment, and quality). I verified solutions on the production line and helped implement scalable improvements for high-volume operations.

• Developed organizational skills through managing multiple tasks simultaneously while adhering to strict deadlines.
• Prepared project presentations and reports to assist senior staff.
• Utilized strong communication abilities during presentations which led to increased understanding among colleagues regarding project goals and objectives.

I designed, assembled, and tested a proton-exchange-membrane (PEM) fuel cell system, covering both core stack integration and supporting subsystems such as cooling, heat exchange, and hydrogen storage tank design. Through iterative prototyping and performance testing, I gained hands-on experience in system-level renewable energy engineering, from component selection and mechanical integration to safety and reliability considerations. I also served as a lecturer for outreach programs, delivering fuel cell and clean-energy workshops and demonstrations at dozens of junior and senior high schools, translating complex engineering concepts into clear, practical learning experiences.