Agentic AI design on AMD Ryzen AI
Join us for our upcoming Future Computing Seminar Series
Speaker: Mario Ruiz (AMD)
Date: October 10th, 2025, 9:30 CET
Where: LEE E 101
Abstract:
Building trustworthy and capable agentic AI systems requires a foundation of open innovation, scalable compute, and transparent design principles. In this talk, we explore how open-source initiatives — with AMD as a key driver — are accelerating the development of autonomous AI agents. We will highlight how AMD’s latest Ryzen™ AI processors, with integrated NPU, provide optimized on-device compute for real-time inferencing and agentic workloads. Additionally, we will dive into Lemonade Server, AMD’s open-source framework designed to simplify and democratize access to AI model deployment and orchestration. Through real-world use cases, we will demonstrate how open software, and collaborative ecosystems are critical for advancing safe, efficient, and adaptable agentic AI architectures. Open-source is no longer optional — it is essential for the future of agent autonomy.
- What is Agentic AI?
- Autonomy, tool use, goals, planning, and memory
- Ryzen AI Overview
- NPU architecture, performance edge, energy efficiency
- Intro to AMD Lemonade Server
- What it is, why it matters (low-latency local inference)
- Open-source deployment flexibility
- What is RAG?
- RAG components: retriever, vector store, generator
- How agents use RAG for context-aware memory
- Tutorial and work thought: Local RAG agent with AMD-optimized model using Lemonade Server
Bio:
Mario is a member of technical staff in the AMD University Program at AMD. As part of this role he delivers training workshops for academics on the latest AMD tools and technologies. He is responsible for managing the Heterogenous Accelerated Compute Cluster globally. Mario has worked on VNx integrating an open source 100 GbE network stack on Alveo platforms for the Vitis flow. He is also part of the PYNQ project – a Python-based open-source productivity environment for Zynq, Zynq MPSoC and Alveo, where he contributed with the Composable Overlays. Mario has over 10 years of experience of designing high performance implementations for FPGAs and SoC. Over the past 5 years, Mario has helped academics and researchers to leverage AMD and Xilinx technologies to achieve the best possible performance/efficiency by providing direct support and delivering training. Before joining AMD, Mario completed his PhD in the Autonomous University of Madrid, which was focused on exploring High Level Synthesis tools in the context of high-speed networking. Mario has a background on electronics, digital design and computer architecture.