GPU Stack Build Architect

Advanced Micro Devices, Inc•San Jose, CA

1d•Hybrid

About The Position

AMD's AI software stack is moving fast — and keeping pace means shipping complete, validated GPU stack releases to customers as quickly as the software can evolve. Today, that release velocity is limited by the coordination overhead between the layers of the stack: firmware, kernel driver, and ROCm each have their own build systems, their own workflows, and no shared baseline. Every release requires manual effort to assemble and validate a coherent recipe across all three. We're building the infrastructure to change that — a unified build platform that treats firmware, kernel, and ROCm as a single deliverable, built from source, validated together, and releasable on demand. The result: customers get supported, compatible GPU stack recipes faster, and AMD can respond to customer needs without the drag of disconnected build and release processes. This is a greenfield architecture role. You'll be making foundational decisions during the proof-of-concept phase, with 1,000+ developers and 65+ firmware components as the eventual scope. If you want to design a build system from scratch — and you reach for an AI coding agent before a bash script — this role was written for you.

Requirements

10+ years of software engineering experience with deep focus on build systems
Strong hands-on coding ability — this is an IC lead role, not a managing-from-above role
Expert-level knowledge of build tools (cmake, ninja, Bazel, or equivalent) and build system design at scale
Experience with packaging, host dependency management, and toolchain configuration
Track record of modernizing or architecting a build system used by 100+ developers
Strong understanding of version control and dependency management at scale (git submodules, manifest-driven workflows, etc.)

Nice To Haves

Fluency with agentic AI workflows (Cursor, Claude, Copilot, etc.) as a force multiplier for engineering throughput
Experience with firmware or kernel build systems (embedded firmware, Linux kernel, or similar)
Familiarity with GitHub Actions and CI/CD pipeline design
Experience building in or migrating to cloud-hosted runner environments (AWS)

Responsibilities

Make foundational build system architecture decisions — the super-build is greenfield. You'll determine how 65+ firmware components, the kernel driver, and ROCm are structured, how dependencies are expressed and resolved, and how the system scales as more components onboard. These decisions matter and they're yours to make.
Lead firmware recipe migration using AI-assisted workflows — the existing firmware builds are scattered across multiple CI systems with no single source of truth. You'll reverse-engineer what exists, understand the dependencies, and convert those recipes into the unified build — using agentic AI coding tools to move at a pace that would otherwise take years. Strong opinions about package management and host dependency handling are a real advantage here.
Build repo automation that keeps the super-repo sane — with 65+ component repos feeding into a unified system, manual dependency updates don't scale. You'll map the full repo landscape, design the submodule/manifest architecture, and build the automation that keeps versions in sync without constant human intervention.
Unblock new team members from day one — the tiger team is actively growing and new engineers are blocked until they have a working dev environment. You'll stand up replicable, documented machine setups that solve the network access, firewall, and cloud quota constraints so onboarding stops being a bottleneck.