• Designed and developed a full billing system used for aggregate all the customers’ cost on our platform.• Responsible for the development and maintenance of an internal distributed object storage system like amazon s3 (100PB) and currently working on designing and developing a new neo storage backend, inspired by Facebook's F4, capable of storing EB-level data at around 15% lower cost than the original version.• Led the development of a file system around the object storage infrastructure, designed to be compatible with POSIX and provide write caching (based on Fuse), along with providing massive small file aggregation indexing, similar to Haystack (Facebook). This system stores and indexes billions of small files to support data supply for AI training, improving the efficiency of expensive hardware utilization during LLM training.• Managed the construction of an internal unified data center, responsible for data governance, and designed and developed an auto-scaling P2P data supply service to enhance the efficiency of data cleaning ETL.• Conducted research on DirectIO and IOUring and successfully applied them to internal storage services, doubling I/O read-write bandwidth.• During the internship at MEGVII, customized and delivered a private version of the platform (including object storage) for over 40 microservices to clients (CI/CD, Operator, Helm, Bazel, dockerize).