Intelligent Infra AI Agent:

Built an autonomous agent using Claude Sonnet (Bedrock) and Strands SDK for multi-account AWS operations and incidents investigation via Slack, using MCP, API skills, memory, etc.

Enterprise Data Lineage & Governance

Architected a data lineage system for full visibility across EMR (Hive, Spark) and Airflow. using Java, Spring Boot, OpenLineage, Kafka, etc.

Designed and implemented Amazon PSPP(Payment Service Provider Program) integration for Zyla, enabling suppliers to receive global payments on Zyla from Amazon.

Implemented identity data-sharing API and file exchange between Amazon and Zyla, using Java, Microservices, RPC, Message Broker, Data Warehouse, SQL, SFTP, etc.

Took part in Transparent data encryption (TDE) database migration to ensure our sensitive customer data is encrypted at rest, using Oceanbase, DRC data synchronization, distributed database, etc.

Successfully migrated several online databases with minimal loss of availability(several minutes).

Implemented automation for a comprehensive quarterly Finance Report, including more than 20 indices and more than 100 numbers, using Java, Spring Framework and JXLS.

Integrated data loading into HTTPS service using Oat++ framework, including C++ and REST API, providing a unified, efficient end-to-end service for loading data into the Vertica database from local files

Simplified the previous data loading workflow from five-step to one-step, highly improved the data loading efficiency

Handled both compressed and uncompressed data files loading inside the single HTTPS service

Designed and built an online knowledge fusion pipeline in Insurance Knowledge Operation Platform using Java, Spring Framework and Serverless Task, consisting of Hospital Standardization, Knowledge Update, Manual Annotation and Bad Case Storage

Implemented Alias Matching, Address Parsing and Geo-location Recall in Hospital Standardization leveraging Microservice and Elasticsearch

Largely reduced hospital data processing cycles -- from 2 days to a few minutes. Highly improved data quality and data processing efficiency, further improved efficiency and accuracy of claims settlement

Detection of Unsealed Muck Trucks: Changed the model to detect all muck trucks and added a new training dataset based on pictures containing big vans which are similar with muck trucks. Improved accuracy and generalization of the detection model in actual scenes, improved the accuracy from 60% to 94% on a new dataset.