AWS Serverless & AI at Scale: From Data Pipelines to LLMs

Join us for an evening focused on building modern AI/ML architectures on AWS. We'll explore two distinct but powerful approaches. First, we'll dive into a serverless medallion architecture through a real-world customer case study, showing how to build a scalable data pipeline using EMR Serverless and AWS AI services like Translate and Textract. Then, we'll shift gears to discuss how to run and scale large language models (LLMs) on Amazon EKS Auto Mode, focusing on the benefits of infrastructure control, cost predictability, and the freedom to run any open-source model.

🗓️ Agenda
18:00 - Arrival
18:15 - Serverless Medallion Architecture: Customer Case Study and Live Demo
18:50 - AI at Scale: Running LLMs with EKS Auto Mode
19:25 - Networking

Serverless Medallion Architecture: Customer Case Study and Live Demo
Join us for a presentation on a serverless medallion architecture implemented for a customer, leveraging AWS EMR Serverless and orchestrated with Step Functions. The case study will explore a multi-layer data pipeline using S3 for temporary storage, with PySpark executed on EMR Serverless. The first three layers incorporate AWS AI tools—Translate, Textract, and Locate—for enhanced data processing, with the final layer utilizing an AI tool for advanced insights.
Following the case study, a live demo will showcase key functionalities of the architecture, highlighting the power and flexibility of serverless infrastructure.
Disclaimer: This presentation focuses on the serverless infrastructure design and implementation, not big data processing.

AI at Scale: Running LLMs with EKS Auto Mode
Running large language models (LLMs) on Amazon EKS offers compelling benefits: full control over your infrastructure, transparent and predictable costs, and the freedom to run any open-source model on your terms. In this session, we'll explore why Kubernetes has become the platform of choice for production LLM deployments and how EKS makes it practical:

• Cost transparency and predictability: Understanding exactly what you're paying for compute, avoiding per-token pricing surprises, and optimizing for your specific usage patterns
• Infrastructure control: Choosing your instance types, managing GPU utilization, and customizing your deployment architecture
• Open-source freedom: Running models like Llama, Mistral, or DeepSeek without vendor lock-in, fine-tuning on your data, and maintaining full ownership
• Production-ready patterns: Deploying with vLLM, handling dynamic scaling, optimizing cold starts, and managing model artifacts
• Simplified operations with EKS Auto Mode: How automation can eliminate infrastructure overhead while preserving control

SPEAKERS

Radu Dobrinescu
Radu Dobrinescu is a Senior Partner Solutions Architect at Amazon Web Services (AWS) with 17 years of technology experience. He specializes in containerized workloads and Kubernetes, helping organizations across Central and Eastern Europe design and deploy scalable infrastructure on AWS. Radu focuses on enabling teams to run AI/ML workloads on Amazon EKS, balancing infrastructure control with operational simplicity to deliver cost-effective, production-ready solutions.

Csaba Szegedi
Csaba Szegedi is a Senior Cloud Architect and AWS Community Builder with over 20 years of experience, currently at SnapSoft in Budapest, Hungary, designing and migrating AWS solutions for clients. As the founder of Syntax Nebula, he delivers hands-on AWS training. An AWS Certified Solutions Architect Professional, Csaba actively shares his expertise through meetups and community events.

Adrián Mezei
Adrian Mezei is a seasoned Cloud Architect and AWS Community Builder who leads the Cloud Migration Service Line at Snapsoft. He has worked with enterprise clients to design and implement AWS infrastructures, focusing on serverless architectures, complex landing zone setups, and scalable networking designs. Adrian is also passionate about Infrastructure as Code, using Terraform to build and automate robust, scalable deployment pipelines. Outside of work, he remains deeply involved in the AWS community, sharing knowledge and mentoring others in their cloud journey.

This event was organized in collaboration with AWS Serverless Budapest and AWS User Group Budapest (Containers and AI/ML).