Software Engineer at Yunjing Technology, on-site at MediaTek

Shun-Wei Wang

王舜緯

Embedded Linux · Software–Hardware Integration · Embedded Systems

I build the software that talks to hardware — embedded Linux, HMI, and the layers underneath. Twelve years of it: AI inference profiling and Yocto builds at MediaTek today, human-machine interfaces at Weintek, production line systems at Garmin, and Android BSP and server firmware at Insyde.

Now

2025.12 — Now

Software Engineer @ Yunjing Technology

On-site at MediaTek · Taiwan

  • Measured AI inference performance on the MT8894 platform and traced the bottlenecks, driving a 15–20% overall performance gain that met the customer's requirement.
  • Automated deployment of the system suspend/resume test environment for the NVIDIA N1X program, cutting setup time from 30 minutes to 15.
  • Maintaining and developing Yocto Project builds — the layers, recipes and toolchains that produce custom embedded Linux images.

Before this: HMI design software at Weintek, seven years at Garmin building the software that runs and inspects production lines, and three years at Insyde on Android board support packages and server BMC firmware.

Full history →

Selected Work

StockRebalance

A desktop tool that works out exactly how many shares to buy or sell to bring a portfolio back to its target weights. Built on Qt and libyahoofinance, for my own financial planning.

C++QtDesktop
github.com/awe31402/StockRebalance

libmodernrobot

A lightweight C++ library for robot arms: the Denavit–Hartenberg model, forward kinematics, and inverse kinematics — small enough to drop into an embedded project.

C++KinematicsRobotics
gitlab.com/awe31402/modern_robotics

linux-interview

Open notes on Linux kernel internals — 340 questions across memory management, scheduling, concurrency and interrupts, plus a practical guide to crash-dump debugging. Written while working through Running Linux Kernel.

Linux KernelARM64crash
github.com/awe31402/linux-interview

Edge AI & model optimization

Where I am pointing the next few years: quantization, pruning, and getting large models to run on small hardware. Course work, paper reading, and ten chapters of runnable code.

Edge AIQuantizationONNX
See the notes →

Taoyuan, Taiwan · Hand-written HTML and CSS, no framework.