1. Work as undergraduate teaching assistant for Introduction to Management Systems Hold office hours and maintain a weekly contribution level of 20 hours to help students better understand management system mechanisms and help gdb debug code. Answering students’ questions2. Design the course project by reading Linux XV6 source code, and related research papers. Explore course topics related to virtual memory, threads, context switch, kernel, interrupts, system calls, inter-process communication, coordination, and interaction.

1. Maintained & Deployed gRPC, DL, and HPC applications on HPC clusters. Monitored clusters through Kubectrl, Kubernetes and Docker. Operated SSH server & container through Linux bash scripts, pipes and Epoll commands.2. Resolved user's ticket related to dataset loading I/O problems in HDFS. Utilized Perf, Netperf, mdtest, top, ioprof and Glances for OS metric collection and further analysis. Identified and rectified backend HDFS I/O issues and centralized metadata management problems.3. Used POSIX - Truncate, Wrap, NFS Mount, Loop Device and 1735+ lines of C code to write a FS middleware that can avoid frequent metadata fetching and wrap small files needed by DL applications in clusters without modifying POSIX and preserving data consistency. 4. Incorporated Redis as middleware for data I/O, adjusted the system architecture in line with SQL HiveSQL and Presto backend Hadoop service system, integrated underlying I/O call methods. Applied SHA-256 hashing to protect user’s privacy and security. 5. Deployed in Tencent HPC Machine and K8S to write commonly used bare metal request processing tasks, enabling pre-calculation of these embedding through scheduled tasks and thus directly transforming the SQL table during common time periods. This improved latency by 1.3X and reduced server CPU usage by 2.25X, as measured by Perf and Kubectrl. Integrated a CLI‐based open-source data pipeline collector, used by 20+ services to process streaming data. Scaled the service to ingest concurrent data streams at gigabit speeds from 4k+ devices, optimized throughput by aging the gRPC/gNMI subscriptions via a tunnel server for load balancing.

1. Designed JD-Buyers-data processing flow with Python and K8S. Work on personalization in JD Buyers App. Develop and maintain Data Storage System, Data Pipeline for JD Recommendation System that record customer tastes regarding rent houses’ information (e.g., Locations). 2. Created algorithm to draw JD transaction Visualization in pipe by Airflow Pipeline and Maps API.Build the JD Buyer API with Spring, Hibernates, CDK, K8S. Optimized JD buyer’s operational monitoring capability and increase service’s availability DAU 10.7X for customers. Implemented distributed caching and load balancing techniques, enhancing operational monitoring capabilities through Prometheus and Grafana integration. Achieved a 10.7X increase in Daily Active Users (DAU), ensuring high availability for a growing customer base. To maintain Buyer’s API, Collaborated with Operations Team, contributing to Docker file and Bash script development, CI/CD pipeline enhancement, and AWS infrastructure utilization including Kubernetes, Kubectl, Perf and Epoll.3. Restructured the GPU cluster k8s scheduler with Golang, including the Kafka parts, contributing over 2127+ LOC. Adjusted Kafka parameters, such as batch.size(for producers) and fetch.max.bytes(for consumers), optimized parallelism, including the channels in Zero Copy's Broke.4. Work Closely with senior machine learning scientist to build JD Buyer’s Recommendation Pipeline. This project helps developers saving 87% of project testing time. By reorganizing airflow pipeline’s structure in the testing phase, optimized ML-Model inference time by 13% and greatly saves the time and storage costs. Utilized API, BASH, SQL Trigger to cache customer taste information. Utilized Redis to store user’s profile and history. Incorporated Redis as middleware for data I/O, adjusted the system architecture and integrated into underlying I/O call methods.

1. Designed online and offline promotion plans using KOC and KOL to attract new users, developed an in-depth understanding of the market industry workflow and corporate organizational structure, screened target users for promotion. Automated daily retrieval of 200+ tables using Selenium from major E-commerce platforms, saving up 83+% spent for data retrieval with RPA tool and winning recognition from VP.2. Assisted the product department with SEO website optimization, developed the company's WeChat accounts in gaining more than 20 million followers, and attained over 30 million article page views with steady increasing retention by 5.4%. 3. Responsible for the user operation of the maintained buy-side, collected user needs and provided timely feedback to the technical department. Assisted PE and Risk Team with integrated Hive SQL on 400+ shoes’ factory policies and exposure data. Consolidated 10-year high frequency transaction data from 100+ Nike retail stores and transformed them into uniformed data structure using Python and SQL. 4. Compiled industry factory insights in 2 presentation decks through actively participating in-site visits and engaging in management meetings, applying communication and interpersonal skills. Identified Chongqing Nike Factory Location with 45% upside potential.