Alan Tseng
I’m an embedded software engineer who uses AI agents to make firmware work go faster.
Rack provisioning at Pegatron, OpenBMC firmware at Wiwynn, Android platform work at MediaTek — three platforms in two years, each with its own toolchain, its own debugger, and its own definition of “it booted”. I don’t think that’s a scattered résumé; I think getting productive on unfamiliar hardware quickly is the skill, and I’ve had to do it repeatedly rather than talk about it.
The deepest register-level work is on the server side: I2C/PMBus drivers, device tree, and cross-machine debugging with gdbserver, live-attach and core dumps, which I presented at COSCUP 2025. On the Android side the leverage has been automation — agent skills that take the repetitive half of platform work off the critical path.
That automation instinct isn’t new. It started with rack-scale provisioning and has since moved up into firmware development itself.
Selected Projects
Talks
| Date | Title | Where |
|---|---|---|
| 2026 | Firmware Development with Agentic AI: Let the AI Run the Build, You Solve the Hard Problems 韌體開發與 Agentic AI：讓 AI 跑 Build，你來解難題 | COSCUP — Taiwan |
| 2025 | Using GDB Debug in OpenBMC | COSCUP — Taiwan |
Experience
Embedded Software Engineer @ MediaTek
MediaTek Corporation, 2025 – Present
Android platform work. Built AI agent skills that automate the repetitive half of the job — pulling Gerrit changes and validating them locally, driving build → flash → boot verification, and walking new-chip porting checklists.
Embedded Software Engineer @ Wiwynn
Wiwynn Corporation, Dec 2024 – Sep 2025 · BMC Firmware
OpenBMC firmware on hyperscaler server platforms — C/C++ drivers over I2C/PMBus, Yocto builds, and the D-Bus/systemd services above them. Debugged production BMCs across machines with gdbserver, live-attach, and core dump analysis, and shared that workflow at COSCUP 2025.
Software Engineer @ Pegatron
Pegatron Corporation, Sep 2024 – Dec 2024 · Server Validation
Automated rack-scale server validation: a Kubernetes-based test system driving PXE OS provisioning across ARM and x86 servers, down at the ARP/DHCP/TFTP layer. Cut the image and repository footprint from 663 MB to 2.4 MB — about 99% — by stripping base images and collapsing redundant layers. This is where the automation thread starts.
Software Engineer Intern @ LINE
LINE Corporation, Jul 2022 – Nov 2022
Built an internal tool that generated type definitions and backend API bindings from Swagger as a single source of truth, cutting API development time by 20%. Deployed it on Kubernetes with DNS and load balancing.