Pang-Yuan Pao

I am Pang-Yuan Pao, an AI Compiler Engineer at MediaTek, focused on building and optimizing compilers for on-device Neural Processing Units (NPUs). With a Master's degree in Multimedia Engineering from National Yang Ming Chiao Tung University (NYCU) and a research background in Computer Vision, I specialize in bridging high-level AI workloads with low-level system optimizations.

Currently, my core work includes:

• NPU Compiler Development: Developing and refining AI compilers for on-device NPUs, specifically implementing hardware-aware optimizations for MediaTek's Deep Learning Accelerator (MDLA).

• Performance Tuning: Profiling performance bottlenecks and developing features like hint-guided option tuners to maximize NPU runtime efficiency and minimize memory footprint.

• System Software & ML Infrastructure: Writing robust, production-level C++ code while leveraging ML frameworks (TensorFlow) and modern compiler technologies (MLIR) to deliver highly efficient, hardware-aware AI solutions.

As a researcher at the Human-centered Intelligent System Lab, my work on intelligent driving technologies has equipped me with a strong, top-down understanding of complex AI workloads. This system-level insight allows me to design highly efficient bottom-up compiler optimizations. Beyond my technical expertise, I am a collaborative problem-solver who thrives in cross-functional environments, consistently driving complex projects to success.

I am passionate about advancing generative AI and edge computing, continuously exploring ways to make on-device AI faster and more efficient.

MediaTek — Hsinchu, Taiwan

AI Compiler Engineer

May. 2025 — Now

EGO INNOVATION — Hualien, Taiwan

Software Engineer

Nov. 2024 — Apr. 2025

HCIS Lab at National Yang Ming Chiao Tung University — Hsinchu, Taiwan

Master Student, Advisor: Yi-Ting Chen 

Sep. 2021 — Jan. 2025

National Yang Ming Chiao Tung University

Master of Multimedia Engineering

Sep. 2021 – Jan. 2025

National Sun Yat-Sen University 

Bachelor of Computer Science and Engineering

Sep. 2017 – Jun. 2021