1. Developed, optimized, and maintained an AI inference service - Improved the inference throughput by up to 50%. - Refactored and optimized the AI inference service code using object-oriented programming and design patterns, successfully reducing the codebase from over 4,000 lines to approximately 2,000 lines. - Developed comprehensive unit tests, achieving an impressive test coverage rate of 93%. These tests played a pivotal role in greatly improving the stability of the code. 2. Developed a monitoring system using Prometheus and Loki to track system resource utilization and AI inference metrics on production servers. 3. Developed an AI model deployment flow using GitLab CI/CD to start inference services on different stage environment servers using Kubernetes. 4. Developed a Kubernetes-in-Kubernetes environment on a development server to simulate the production environment across various factory servers. 5. Optimized, and maintained the Auto-Scheduler API for a web training platform. 6. Developed, optimized, and maintained a defect detection model - Capacitor component defect detection model with 0% leakage and overkill less than 1%. - Developed an inference model and implemented the corresponding code for detecting defects in IC components.

1. Build deep learning training APIs Developed and integrated deep learning training APIs, enabling the implementation of diverse algorithms, such as object detection, object segmentation, classification, anomaly detection, and OCR. These APIs were instrumental in constructing a user-friendly, no-code deep learning training platform on a website. 2. Build C++ deep learning inference APIs Developed C++ deep learning inference APIs to convert PyTorch models trained on our platform into C++-compatible models (torchlib). This facilitated the deployment of models on edge computing servers, enabling real-time defect prediction on production lines. 3. Developed and optimized algorithms for defect detection