AMD
From AMD's careers page
Status: listed on the employer's own careers page when we last checked (daily). Confirm on their site before applying.
THE ROLE: AI applications no longer run on a single model.
ADVANCE YOUR CAREER. ADVANCE THE WORLD.
At AMD, we believe technology has the power to solve the world’s most important challenges. From advancing healthcare and scientific discovery to powering AI and the technologies people rely on every day, innovation at AMD is shaping the future.
Whether you’re designing next-gen processors, enabling AI breakthroughs, or bringing leading edge products to market, every role at AMD contributes to something bigger — technology that moves the world forward. Join us and, together, we’ll advance your career.
AI applications no longer run on a single model. They combine frontier and open models, large and small, with agents and tools. Every request needs a decision: which model should answer, at what cost, and under which safety and privacy policies. We build the layer that makes that decision.
AMD is looking for a software engineer to join the core team behind vllm-sr. The team owns the whole stack: the router, the serving engine, the decision models and the open-source community. You will train and serve on AMD Instinct GPUs, help make ROCm a first-class platform for intelligent routing, and ship your work upstream as code, open models and papers that reach users around the world.
You are a hands-on engineer who wants to build software that people rely on. You have strong fundamentals in systems, machine learning, or both, and you understand how large language models are trained and served. AI is part of how you work every day, and you have taken products from idea to production with it. You measure before you optimize, and you write code that others are glad to review and maintain.
You are comfortable working in the open, where designs, code reviews and discussions happen on GitHub with a global community of contributors. The work spans systems, models and community, and we don't expect anyone to bring all of it on day one. We look for real depth in one area, such as routing and serving systems, inference performance or model training, and the curiosity to grow into the rest.
Build the routing intelligence of vllm-sr, from decision models that pick the right model for every request to semantic, policy-based and workload-aware routing, semantic caching, and routing for safety, privacy, tools and agents.
Make vllm-sr simple to deploy and operate in production, with first-class integrations for vLLM, Kubernetes, Envoy, Gateway API and AI gateways.
Build vllm-sr into a universal serving engine for decision models that exposes the System One API and runs efficiently on everything from AMD Instinct GPUs to CPUs.
Optimize decision model inference across batching, memory, latency and throughput, down to GPU kernels in HIP or Triton on ROCm where it matters.
Train our foundation decision models across sizes and keep pushing their capabilities to the frontier, owning data, training, post-training and evaluation.
Build an auto research pipeline that keeps iterating on our decision models, automating experiments from data generation to training, evaluation and analysis.
Co-design the decision models with the serving engine, so that every new capability ships with the performance to serve it.
Make decisions with data: build benchmarks and evaluations, profile end-to-end systems, and measure routing quality, latency and cost.
Ship production-quality software with strong testing, CI, observability and reliability.
Work upstream first: review and land community contributions, triage issues, ship releases and help the community grow.
Represent AMD and vllm-sr at conferences and workshops, and share the work through talks, demos, blogs and documentation that grow the community and show what vllm-sr can do on AMD GPUs.
Partner across AMD: with ROCm, kernel and AI framework teams to remove bottlenecks and validate vllm-sr on new AMD GPU platforms, and with Enterprise and IT teams to support AMD's internal products and IT services with vllm-sr.
Fresh Graduates with strong programming skills in Python and at least one of Go, Rust or C++.
Hands-on experience with large language models: serving them (vLLM, SGLang or similar), training them, or building products on them.
AI-native engineering: deep, daily use of AI coding agents and agentic workflows to design, build, test and ship, with the judgment to verify and correct their output.
Proven end-to-end delivery: products you built and shipped with AI driving the work, from idea and design through deployment and operations.
Solid fundamentals in systems, machine learning, or both.
Comfortable working in the open: GitHub workflows, public code review and clear written English.
Contributions to open-source projects such as vllm-sr, vLLM, SGLang, PyTorch, Hugging Face, Envoy or Kubernetes.
GPU programming experience (HIP, CUDA, Triton), or ROCm on AMD Instinct GPUs.
Experience with model routing, API gateways or LLM inference optimization.
Experience training or post-training models such as classifiers, reward models or multimodal models.
Seoul, Korea
Benefits offered are described: AMD benefits at a glance.
AMD does not accept unsolicited resumes from headhunters, recruitment agencies, or fee-based recruitment services. AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition, mental or physical disability, national origin, race, religion, political and/or third-party affiliation, sex, pregnancy, sexual orientation, gender identity, military or veteran status, or any other characteristic protected by law. We encourage applications from all qualified candidates and will accommodate applicants’ needs under the respective laws throughout all stages of the recruitment and selection process.
AMD may use Artificial Intelligence to help screen, assess or select applicants for this position. AMD’s “Responsible AI Policy” is available here.
This posting is for an existing vacancy.
AI applications no longer run on a single model. They combine frontier and open models, large and small, with agents and tools. Every request needs a decision: which model should answer, at what cost, and under which safety and privacy policies. We build the layer that makes that decision.
AMD is looking for a software engineer to join the core team behind vllm-sr. The team owns the whole stack: the router, the serving engine, the decision models and the open-source community. You will train and serve on AMD Instinct GPUs, help make ROCm a first-class platform for intelligent routing, and ship your work upstream as code, open models and papers that reach users around the world.
You are a hands-on engineer who wants to build software that people rely on. You have strong fundamentals in systems, machine learning, or both, and you understand how large language models are trained and served. AI is part of how you work every day, and you have taken products from idea to production with it. You measure before you optimize, and you write code that others are glad to review and maintain.
You are comfortable working in the open, where designs, code reviews and discussions happen on GitHub with a global community of contributors. The work spans systems, models and community, and we don't expect anyone to bring all of it on day one. We look for real depth in one area, such as routing and serving systems, inference performance or model training, and the curiosity to grow into the rest.
Build the routing intelligence of vllm-sr, from decision models that pick the right model for every request to semantic, policy-based and workload-aware routing, semantic caching, and routing for safety, privacy, tools and agents.
Make vllm-sr simple to deploy and operate in production, with first-class integrations for vLLM, Kubernetes, Envoy, Gateway API and AI gateways.
Build vllm-sr into a universal serving engine for decision models that exposes the System One API and runs efficiently on everything from AMD Instinct GPUs to CPUs.
Optimize decision model inference across batching, memory, latency and throughput, down to GPU kernels in HIP or Triton on ROCm where it matters.
Train our foundation decision models across sizes and keep pushing their capabilities to the frontier, owning data, training, post-training and evaluation.
Build an auto research pipeline that keeps iterating on our decision models, automating experiments from data generation to training, evaluation and analysis.
Co-design the decision models with the serving engine, so that every new capability ships with the performance to serve it.
Make decisions with data: build benchmarks and evaluations, profile end-to-end systems, and measure routing quality, latency and cost.
Ship production-quality software with strong testing, CI, observability and reliability.
Work upstream first: review and land community contributions, triage issues, ship releases and help the community grow.
Represent AMD and vllm-sr at conferences and workshops, and share the work through talks, demos, blogs and documentation that grow the community and show what vllm-sr can do on AMD GPUs.
Partner across AMD: with ROCm, kernel and AI framework teams to remove bottlenecks and validate vllm-sr on new AMD GPU platforms, and with Enterprise and IT teams to support AMD's internal products and IT services with vllm-sr.
Fresh Graduates with strong programming skills in Python and at least one of Go, Rust or C++.
Hands-on experience with large language models: serving them (vLLM, SGLang or similar), training them, or building products on them.
AI-native engineering: deep, daily use of AI coding agents and agentic workflows to design, build, test and ship, with the judgment to verify and correct their output.
Proven end-to-end delivery: products you built and shipped with AI driving the work, from idea and design through deployment and operations.
Solid fundamentals in systems, machine learning, or both.
Comfortable working in the open: GitHub workflows, public code review and clear written English.
Contributions to open-source projects such as vllm-sr, vLLM, SGLang, PyTorch, Hugging Face, Envoy or Kubernetes.
GPU programming experience (HIP, CUDA, Triton), or ROCm on AMD Instinct GPUs.
Experience with model routing, API gateways or LLM inference optimization.
Experience training or post-training models such as classifiers, reward models or multimodal models.
Seoul, Korea
Benefits offered are described: AMD benefits at a glance.
AMD does not accept unsolicited resumes from headhunters, recruitment agencies, or fee-based recruitment services. AMD and its subsidiaries are equal opportunity, inclusive employers and will consider all applicants without regard to age, ancestry, color, marital status, medical condition, mental or physical disability, national origin, race, religion, political and/or third-party affiliation, sex, pregnancy, sexual orientation, gender identity, military or veteran status, or any other characteristic protected by law. We encourage applications from all qualified candidates and will accommodate applicants’ needs under the respective laws throughout all stages of the recruitment and selection process.
AMD may use Artificial Intelligence to help screen, assess or select applicants for this position. AMD’s “Responsible AI Policy” is available here.
This posting is for an existing vacancy.
Scores your resume against this job the way resume scanners do: the job's hard and soft skills, searchability, recruiter tips and formatting.
Create a free account to see your score. Your resume stays in this browser until then, and we never store it.
Already a member? Sign in