Chen Yang

Senior Software Engineer, May 2021 - Present.

• Mentored interns, and supported cross-team technical discussions and planning.
• Coordinated with cross-functional teams to acquire additional testing resources, improve regression efficiency, and enhance test coverage.
• Continuously fixed broken tests, updated infrastructure to support new environments, and drove migration efforts. Introduced tools to improve test stability, reducing wait times, and increasing test reliability.
• Actively worked on resolving development pain points by researching and experimenting with testing tools, while addressing critical issues, and providing ongoing support for runtime testing infrastructure.
• Led initiatives to improve development velocity by streamlining runtime stack separation, gathering stakeholder feedback, and documenting performance impact.
• Refined Slurm and Kubernetes integration by fixing critical issues, enabling container support, enhancing container logging, and meeting customer-specific deployment needs.
• Maintained and enhanced a discrete event simulation tool for modeling graph execution and data transfers; improved automation, streamlined the pipeline, and updated profiling methods to better support hardware scaling and performance analysis, while driving adoption through a company-wide tech talk.
• Improved DDR to HBM checkpoint efficiency of the LLM model by 60% (1,900 ms to 780 ms) via runtime profiling and data structure optimizations.
• Added key tensor conversion support in collaboration with the compiler team.
• Championed runtime infrastructure improvements, significantly reducing regression failures.
• Added multi-context support for the debugger to aid in troubleshooting of distributed workloads on the same system.
• Evaluated and analyzed the runtime performance of the deep learning model (BraggNN) in a cross-functional project, and helped to improve the runtime performance of training by 50%.
• Prototyped the first pipeline-parallel ML model across multiple SambaNova systems, validating a key Composition of Experts capability.

Software Engineer, Feb 2020 - May 2021

• Implemented interactive GDB-like breakpointing functionalities for the novel Sambanova accelerator debugger.
• Developed device plugins to facilitate allocating, monitoring, configuring, and scheduling Sambanova accelerator resources for resource allocation and scheduling systems like Slurm (HPC) and Kubernetes (container orchestration).

• Created 3D point cloud map for sidewalk autonomous driving delivery robot to use in localization.
• Built data conversion tool in C++ for custom robot system to convert collected sensor data (Lidar, IMU, odometry) during actual test drives into Rosbag files for analysis and mapping in the Robot Operating System.
• Finished parameter tuning to produce a 3D point cloud map of a 5km by 5km urban region in Tokyo visualized in RVIZ using google's cartographer package in the Robot Operating System.

• Company Overview: Funded by Boeing and Foxconn
• Collaborated with a PhD student and staff engineers to enable the Mobile Platform to detect foreign object debris inside wings of Boeing airplanes using a modular manipulator with series elastic actuators.
• Designed and completed experiments to calibrate custom designed end-effector force sensor.
• Implemented teach pendant software using Matlab with end-effector force sensing capability on a 8-degrees-of-freedom robotic arm.
• Created a 6-degrees-of-freedom space-mouse control interface in Matlab for the 8-degree-of-freedom robotic arm.
• Partnered with two staff engineers to allow orientation data from IMU to be fused with ultra-wide band localization module.
• Deployed robot in experimental factory setting and demoed localization result with real-time visualization in RVIZ.
• Implemented communication between IMR's localization/vision embedded module and other modules through ROS(Robotic Operating System) and ROS-Serial with USB-serial/Ethernet options in Python.
• Utilized the Mahony filter to perform IMU 9-axis sensor fusion on Arduino and produced quaternion heading result.
• Funded by Boeing and Foxconn