• Designed a troubleshooting system for the science team to pinpoint bugs in the brand infringement analysis process, boosting the precision and dependability of brand detection models, reducing data retrieval time by at least 25%. • Developed a Java service utilizing AWS components (Lambda, SQS, SNS, Step Functions, DynamoDB) to automate the collection and processing of training data, including manual suppression, prediction outputs, and suppression rules. • Improved the frontend of an internal tool with React, adding intelligent access buttons that auto-disable when inactive and dynamically link to relevant URLs, enhancing usability and collaboration. • Configured dynamic scaling in AWS SageMaker to automatically adjust endpoint targets based on traffic demands. • Created extensive integration and unit tests to ensure the long-term stability and reliability of implemented features.