Alexander Zhang

Work Experience

• Responsible for C++ inference pipeline across all image apps
• Converted models from research-grade Pytorch to production ONNX Runtime, TensorRT, CoreML, and OpenVINO for better performance and broad desktop support
• Wrote platform specific inference code for diverse CPU/GPU/NPU architectures
• Worked with Nvidia/AMD/Apple/Intel/Qualcomm to improve speed and resource usage
  • Doubled upscale model throughput on Snapdragon through GPU & NPU optimizations
  • Reduced autopilot latency on 24MP images from 1.8s to 0.9s on Linux L40S
• Shipped Gigapixel 5 & 6, with face recovery, viewport caching, etc. as solo dev

• Designed and implemented cooperative multitasking OS for Cortex-M3 based WSN
• Developed LTE modem driver, OTA update system, and HTTP request parser

• Processed and analyzed regional climate model data from NCAR supercomputers

Education

• B.S.E. in Computer Science

Selected Projects

• Nintendo 64 emulator for Windows/MacOS that runs many commercial games
• Wrote MIPS to x64 JITs for both CPU and RCP, with SIMD to SIMD emulation
• Emulated RDP using software rasterization in Vulkan megakernel/ubershader

• Wrote LLVM backend and newlib toolchain ports for novel RISC ISA
• Implemented CPU, DDR3 controller, display, etc. on Intel FPGA
• Ported Doom, Coremark, and other software to the platform

• Native macOS implementation of Stable Diffusion, predating Apple release
• Converted model to CoreML and optimized memory reuse for Intel Macs
• Implemented tokenization, scheduling, and noise generation in Objective-C

• Webapp for making mockups of custom keyboard layouts and keycap set designs
• Widely adopted by mechanical keyboard community, >200 renders daily in 2019

Technical Skills

• Languages: C, C++, Python, x86-64/AArch64/MIPS, HLSL
• Inference: ONNX Runtime, TensorRT, DirectML, CoreML, OpenVINO, QNN, Pytorch
• Image/Video: OpenCV, OpenImageIO, ffmpeg/libav, CUDA, Vulkan
• Platforms: Windows, macOS, Linux, iOS, embedded systems