Forward Deployed Engineer (FDE)

About the Role We are seeking a Member of Technical Staff to join our Forward Deployed Engineering (FDE) team. In this customer-facing role, you will work directly with strategic partners and customers as part of the FDE team to enable, optimize, and deploy state-of-the-art GenAI inference workloads on AMD data center CPUs and GPUs (Instinct family). You will bridge the gap between cutting-edge AI models and AMD hardware, driving competitive performance across the full inference stack. Responsibilities • Partner with customers as part of the FDE team to enable and optimize large language models (LLMs) and multimodal models on AMD Instinct GPUs, meeting committed performance and accuracy targets. • Optimize GenAI inference pipelines on vLLM and SGLang, including PD disaggregation, expert parallelism (EP/wide-EP), tensor/data parallelism, speculative decoding, and KV cache management. • Develop and tune low-level GPU kernels (Triton, HIP, composable kernels) for attention, MoE, GEMM, quantization, and normalization operations. • Profile end-to-end workloads, identify bottlenecks across GPU/CPU boundaries, and deliver measurable throughput and latency improvements versus competitive targets. • Enable advanced data types and techniques (FP8, MXFP4, sparse attention) with a focus on both performance and numerical accuracy. • Debug complex runtime, stability, and accuracy issues in production environments; partner with core engineering teams to identify root causes and contribute fixes. • Contribute optimizations back to open-source frameworks (vLLM, SGLang, ATOM, AITER) and internal libraries; document best practices and deliver production-ready artifacts to customers. • Leverage AI-assisted coding tools and agentic workflows to accelerate development, debugging, optimization, and code review across the inference stack. What You'll Bring • BS, MS, or PhD in Computer Science, Electrical Engineering, or a related field, along with 8+ years of relevant experience. • Solid expertise in GPU architecture and performance optimization, including hands-on experience with AMD ROCm (or CUDA) and GPU kernel development. • Strong proficiency in Python and C++, with experience developing and optimizing kernels using Triton, HIP, or CUDA. • Experience with AI-assisted coding and agentic development tools (e.g., GitHub Copilot, Claude, Cursor, or equivalent) to enhance productivity in kernel development, large-scale codebase navigation, debugging, and framework integration. • Experience working with LLM inference frameworks (vLLM, SGLang, or equivalent) and serving optimization techniques. • Solid understanding of transformer and MoE architectures, quantization, and distributed inference strategies (TP, DP, EP, and related parallelism techniques). • Proficiency with profiling and debugging tools (e.g., rocProfiler, ROCdebug-agent). Nice to Have • Experience optimizing frontier models (DeepSeek, Qwen, Kimi, MiniMax, GLM, etc.) in production environments. • Track record of contributing to open-source AI inference frameworks. • Experience integrating AI-driven automation into engineering workflows and developer tooling. • Strong customer-facing communication skills and the ability to manage technical deliverables against committed timelines. • BS, MS, or PhD in Computer Science, Electrical Engineering, or a related field, along with 8+ years of relevant experience. • Solid expertise in GPU architecture and performance optimization, including hands-on experience with AMD ROCm (or CUDA) and GPU kernel development. • Strong proficiency in Python and C++, with experience developing and optimizing kernels using Triton, HIP, or CUDA. • Experience with AI-assisted coding and agentic development tools (e.g., GitHub Copilot, Claude, Cursor, or equivalent) to enhance productivity in kernel development, large-scale codebase navigation, debugging, and framework integration. • Experience working with LLM inference frameworks (vLLM, SGLang, or equivalent) and serving optimization techniques. • Solid understanding of transformer and MoE architectures, quantization, and distributed inference strategies (TP, DP, EP, and related parallelism techniques). • Proficiency with profiling and debugging tools (e.g., rocProfiler, ROCdebug-agent).