· Designed and implemented write sync mechanism to sync UPS DDB with CIDS DDB asynchronously through SQS, DLQ and Lambda. Came up three system structure to handle the partial records update failure scenario for one SNS message. For non-order and duplicate SNS message, proposed two solutions to deduplicate records and only save latest records.· Monitored performance and traffic of service through Cloud Watch. Emitted customized metrics using Declarative Coral Metrics and Declarative Lambda Metrics. Created dashboard, alarms and monitors based on CDK Construct and TypeScript.· Corrected false alarm by preventing reporting downstream client errors as internal service errors. Improved service availability from ~80% to 100% and resolved 10+ relevant high-severity tickets.Constructed APIs with the Smithy model and Coral framework. Implemented unit tests by Mockito, and integration tests by TestNG and AssertJ. Interacted with AWS Dynamo DB to save or query data.· Created Cradle job, ETL job, and SQL query to analyze data in Andes table. Used Weblab to test new interest's performance.· Manual test server locally using Coral Explorer in Linux Cloud Desktop.

· Captured phrase similarity by implementing BERT and deep Siamese Bidirectional LSTM in Linux to increase accuracy by 20% to compare news and sentence similarity.· Composed remote data transmission port using Python and Flask and encoded files through base64 encoding to transfer the checked news, its similarity, and the spot of similar sentences to the remote end.· Reduced data processing time by 50% through parallel scheduling and model reconstruction with batch method.

· Accomplished a text clustering system with Sequential model and Keras in Python. Improved accuracy to 92.76% using Word2vec algorithm and LSTM automatic coder to extract features, compress code, and generate word vector space.· Built responsive web UI on Bootstrap to present outcomes. Supported web services with Flask, Python, and SQL Server. · Analyzed the feature vector by using PCA to descend its dimension and using Matplotlib to draw a scatter plot.

· Developed web crawler using Beautiful Soup library in Python to mine about 80,000 comments and its number of approvals.· Filtered unimportant words by calculating value of TF-IDF, and removed special symbols through regular matching.· Designed comment sentiment classifier by Word2vec, Doc2vec, and K-means algorithm to distinguish positive reviews and negative reviews.