ML Systems Research Engineer, RL / Inference / Agent Systems at Advanced Micro Devices in Santa Clara, CA

Job description

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. THE ROLE: We are hiring ML Systems Research Engineers to build the reinforcement learning, inference, and evaluation infrastructure behind AI-for-engineering systems. This role focuses on the systems that let agents and models improve real engineering workflows: running many attempts, evaluating correctness, measuring performance, managing long-latency rewards, and feeding results back into model and agent improvement. You will work across compute optimization, hardware engineering automation, verification, simulation, debugging. The emphasis is on scalable ML systems that make research practical, repeatable, and useful for production engineering teams. THE PERSON: You are a systems-minded ML engineer or researcher who understands that model quality depends on the surrounding loop: data, tools, inference, graders, reward design, logging, and iteration speed. You can build reliable infrastructure, reason about RL and inference tradeoffs, and collaborate with scientists and applied engineers to make experiments reproducible and useful. KEY RESPONSIBILITIES Build RL and inference systems for agentic engineering workflows, including job orchestration, sampling, scoring, caching, experiment tracking, and reproducible evaluation. Develop infrastructure for long-horizon and high-latency reward tasks where validation can take minutes to hours. Design staged rewards, proxy graders, sliced evaluation paths, retry strategies, and uncertainty-aware evaluation methods. Support optimization workflows with systems for candidate generation, benchmark execution, correctness checking, profiler feedback, reward modeling, and model-level improvement. Partner with AI research scientists on reward hacking research, reward shaping, metareasoning, and post-training methods for engineering tasks. Build scalable inference and tool-use pipelines for LLM agents that interact with compilers, profilers, simulators, formal tools, benchmark harnesses, and internal knowledge sources. Standardize datasets, eval definitions, run logs, leaderboards, failure taxonomies, and data collection for future training. Analyze experimental results and turn system behavior into actionable guidance for model, agent, tool, and reward improvements. TECHNICAL FOCUS AREAS Reinforcement learning and post-training infrastructure for tool-using agents. Inference systems for LLMs and agents, including latency, throughput, batching, sampling, reliability, and observability. Evaluation systems for tasks with expensive, delayed, mixed, or sparse rewards. Reward design for engineering domains where correctness, performance, quality, and resource usage must be balanced. Distributed experimentation, job orchestration, caching, data pipelines, dashboards, and reproducible run management. Integration with external tools such as compilers, profilers, simulators, validation systems, benchmark harnesses, and ticketing or knowledge systems. PREFERRED QUALIFICATIONS Strong programming skills in Python and experience with ML frameworks such as PyTorch, JAX, TensorFlow, or similar. Experience building ML systems, RL infrastructure, inference services, agent frameworks, evaluation platforms, or distributed experimentation systems. Strong understanding of model inference, batching, sampling, latency, throughput, observability, and reliability tradeoffs. Ability to design experiments and evaluation pipelines with clear metrics, logs, reproducibility, and statisti