Developed a OCR algorithm optimized for large-scale insurance document datasets Developed a multi-modal document understanding LLM Model by integrating layout-aware embeddings with fine-tuned LLama models, combining visual features from OCR with spatial and textual embeddings to enhance LLM's document comprehension capabilities.Developed an end-to-end insurance AI system capable of processing various document templates, significantly streamlining claims handling and reducing manual intervention, achieving over 90%

Supervisor: Hsu Jia-Lien Research domain: Object detection , Federated learning, LLM, OCR , Document Understading

Containerization and Virtualization:Leveraged Docker to ensure consistent environments across development, testing, and production for applications.System Administration:Managed and optimized LAMP (Linux, Apache, MySQL, PHP) stack configurations to enhance system efficiency and stability.Database Management:Designed and implemented robust database backup solutions across multiple systems, ensuring data reliability and quick recovery in case of system failure.Web Management:Administered servers using Webmin, improving user interface and accessibility for Unix system operations.

- Collaborate with a Vietnamese software development team to ensure comprehensive coverage of manual testing processes for software products.- Writing documentation including user stories, user requirement specifications, and software development specifications aligned with ISO 9001 standards.- Develop and execute detailed test cases and test plans to identify software defects and ensure compliance with the end-user requirements.- Actively participate in the software development lifecycle, applying ISO 9001 workflows to improve product quality and enhance process efficiency.- Provide feedback and detailed reports on testing outcomes to facilitate bug fixes and product enhancements in collaboration with development teams.

* Developed a high-throughput event-driven platform for real-time data feeds and TCP/IP tracking of 10,000+ devices.* Leveraged Apache Kafka to implement a robust Event-Driven Architecture (EDA)* Decoupled, scalable microservices for improved system flexibility and maintainability* Utilized Java Socket.IO for efficient bi-directional communication between server and devices.* Optimized database queries for handling large-scale inserts, achieving a 5x improvement in insert speeds, handling up to 10,000inserts per second.* Utilized SQL and NoSQL databases to efficiently handle different types of data