Home - Zhongqian Duan

Working on the Amazon S3 Glacier team.
Redesigned Free Zone Allocation strategy in Kotlin, a core component for S3 Glacier customer retrievals, standardizing read throughput and reducing tail latency to improve fairness, scalability, and overall system reliability. Validated with comprehensive unit, plugin, and reference tests in production.
Built a standalone storage service on EC2 to validate SCSI read/write operations, enabling Glacier deployment on hardware in regions without legacy network support. This milestone expanded global availability, reduced dependency on legacy infrastructure, and ensured reliable data access for all customers.
Developed an automated operational service to generate CloudWatch queries, retrieve logs, and post results directly to tickets, reducing on-call issue resolution time by 30% and improving debugging efficiency.
Implemented secure, low-latency serverless workflows using AWS Lambda, S3, DynamoDB, and IAM, supporting production-scale workloads with strong data integrity guarantees.

Worked on the UGC > Logic team for the virtual reality gaming startup.
Developed 20+ production-ready logic “chips” in C# for the CircuitV2 visual programming system, enabling dynamic and interactive VR experiences for creators.
Designed and implemented a robust Undo/Redo system for maker-pen wiring, ensuring deterministic restoration of circuit connections and providing maintainable API specifications.
Created the Motion Trail prefab tool using Protobuf (serialization) and Photon Networking (real-time sync), enriching the visual experience of moving objects in multiplayer VR.
Leveraged CI/CD workflows Azure DevOps, and managed tasks in Jira, enabling efficient collaboration and consistent release quality.
Deploy builds on Steam using Jenkins, and collaborated with a team of developers and designers to playtest new features in both screen and VR mode.

Built distributed training infrastructure using Python and gRPC, enabling multi-GPU training, remote validation, and centralized data serving, increasing data throughput by 20%.
Developed a containerized simulation data pipeline with Docker and S3, supporting real-time reinforcement learning workflows.
Trained a foundation model for FEA simulations in PyTorch, achieving a 5% reduction in Mises Stress Loss and faster convergence on general simulation tasks.

Participated in the NIO black-box project that tests the entire autonomous driving system with on-screen video.
Participated in the design and optimization of a 3D Object Tracking Network for autopilot, improved the precision by 4%.
Utilized pre-trained vision models, such as MaskRCNN, with different backbones to detect vehicles and lane lines, and tested for autonomous driving systems on over 2000 on-screen videos.
Proposed a light-weight CNN with different backbones in PyTorch to remove the lens distortion effect.

EECS 587 Parallel Computing Course GSI
Led weekly discussion sessions and office hours for EECS 587 Parallel Computing. (CUDA, OpenMP, MPI)

Zhongqian Duan
SDE @ AWS | M.S. @ UMich (CSE)