Agenda

Abstract

During my time at school, I was amazed by the buildings constructed by different civilizations, their similarities and, at the same time, their uniqueness. Depending on whether it was a house, a farm, a fortress, etc., the planning varied, as it required not only its own design plans but also varied in location, the materials and machinery needed, the number and type of workers (builders, electricians, plumbers, specialists), among other things, and with all this we could then have an estimate of the time and cost of the work.

Does this sound familiar? In this session, I want to share my experience working with different types of applications. I have been fortunate (and old enough) to work with applications ranging from mainframe systems to new multi-agent use cases. We will see how to leverage services such as EC2, ECS, EKS, Lambda, and Bedrock AgentCore across different architectural types.

Bio

I'm a computer engineer and a developer at my core, so I would rather work on hands-on tasks and delivering results in the projects than making Power Point presentations. With experience spanning across multiple industries and countries, my focus lies in the Cloud and Big Data domains, approaching them with a DevSecOps mindset. Additionally, I am a member of the AWS Community Builders program, contributing to the community's growth.

Abstract

Managing cloud cost visibility and allocation across an Enterprise organization with over 50 AWS accounts, multiple Business Units, and shared resources is a complex challenge. Without granular tracking for each user or BU, calculating ROI and optimizing budgets quickly becomes a nightmare.

In this session, I will share a real-world case study: how to transform cost chaos into pinpoint visibility using a two-tier automated tagging strategy designed and executed end-to-end.

**What you will learn in this session:**

* **Two-Tier Tagging Matrix:** How to structure Vertical (Core Infrastructure, Backup/DR, Security) and Horizontal (Business Unit, Environments) tagging.

* **Cross-Account Automation:** How to orchestrate tag enforcement and distribution at enterprise scale using AWS Systems Manager and CloudFormation StackSets.

* **Data-Driven Decisions:** Converting granular tag data into real-time dashboards and automated cost-per-end-user calculations.

Bio

Giovanni Adinolfi is a Senior Cloud Architect and Cloud FinOps Expert with over 25 years of IT infrastructure experience, including more than a decade specializing in AWS ecosystems. A 6x AWS Certified professional holding the Solutions Architect - Professional designation alongside Security and Advanced Networking Specialties, he specializes in multi-account governance, enterprise landing zones, and cloud cost optimization. Giovanni's hands-on track record includes executing FinOps transformations across complex enterprise environments with 50+ AWS accounts. He recently delivered €38K in verified savings on an ~€800K annual cloud budget by building CUR/Athena data pipelines, CUDOS dashboards, and deploying cross-account automated tagging strategies using AWS Systems Manager and CloudFormation StackSets. Beyond cost allocation, he has led high-stakes cloud migrations and architectural advisory for regulated environments, including healthcare platforms. Giovanni excels at bridging the gap between Finance and Engineering leadership, turning complex cloud telemetry into actionable, data-driven decisions.

Abstract

AI agents are becoming increasingly capable of reasoning, using tools and executing complex workflows. But there is still a fundamental problem: how does an agent understand what data is available, what it represents, and which information is actually relevant to its goal?

Traditionally, answering these questions requires additional indexes, databases, metadata catalogs or retrieval layers built around the original data. With Amazon S3 Annotations, AWS is introducing a different approach: bringing rich, mutable and machine-readable context directly alongside the objects stored in S3.

In this session we'll explore how this new capability can change the way we design agentic architectures on AWS. Starting from a traditional data lake or document repository, we'll see how annotations can enrich objects with business context, AI-generated information, classifications and processing results, allowing agents to discover and understand data more dynamically.

We'll discuss how S3 Annotations, S3 Metadata and AI agents can work together to transform S3 from a passive object store into an active part of the agent's reasoning and data-discovery process — moving from "retrieve this file" to "find the information that helps me accomplish this goal."

Bio

I’m Francesco Gallo, a 24-year-old GenAI Engineer at ReCube, an AWS Partner company where I’ve been working for about a year and a half, designing and developing Generative AI solutions built entirely on AWS. I graduated with honors in Computer Science from the "Università della Calabria" and I work daily with AWS services, with a strong focus on Generative AI, agentic architectures and cloud-native solutions. Alongside my professional activity, I’m an active member of the AWS User Group Calabria, where I contribute to the growth of the local AWS community and regularly share my experience through technical talks and community events.

Abstract

Amazon Aurora DSQL is a serverless distributed SQL database with virtually unlimited scale, highest availability, and zero infrastructure management. In this talk we’ll use a sample serverless application that uses Amazon API Gateway, AWS Lambda and Aurora DSQL to explore the architecture and features of this database.

We'll explore the scaling behaviour of DSQL, measure performance including cold starts, deep dive into its pricing and current limitations.

We'll wrap up with some suggestions when currently to use DSQL comparing to DynamoDB, RDS and Aurora (Serverless) and what to pay attention to when migrating your existing application to Aurora DSQL.

Bio

I'm Vadym Kazulkin, AWS Serverless Hero and Head of Development at ip.labs based in Bonn. I have been in the tech industry for over 25 years. My current focus and interests include the design and implementation of highly scalable and available applications in the AWS Cloud, with a special passion for Serverless, Generative, and Agentic AI, as well as the optimization of Java applications (also utilizing Spring Boot, Micronaut, and Quarkus frameworks) on AWS Lambda. I'm also the co-organizer of the Java User Group Bonn meetup, a frequent speaker at various national and international events, and a regular author of articles in different published and online magazines and on my online blog

Abstract

During an active production incident, the AI agent we had built reported the affected system as operating within normal parameters. The runbooks were correct. The historical match was correct. But nothing in our knowledge base had a mechanism for "what is happening right now." A vector index has no answer for it. Even if the knowledge base was correct, it was answering a different question.

We improved the knowledge base. Of course we did. Better retrieval, more documents, higher relevance scores. It did not help. A knowledge base answers one type of question. What do the documents say. A production agent in an operational context needs different answers. What is the current system state. What are the dependency relationships. What happened historically. Each context type has a database built for it, with a different data structure and a different access pattern.

This session shows how to apply purpose-built databases to AI agents. Multiple stores, different context types, each answering what the others cannot.

Bio

I have been building with AWS for ten years and teaching it for seven. In Italy I have worked as a licensed freelance engineer, consultant, and founder. I am now CTO and Co-founder, an AWS Authorized Instructor at Champion level, and an AWS Community Builder for seven years. I design cloud architectures with a focus on AI agent systems, data engineering, and serverless patterns on AWS, and I train thousands of cloud practitioners across Europe on the same architectures I build in production.

Abstract

A community session to play AWS Builder Cards.

Abstract

Dalla ricerca vettoriale con Amazon Aurora PostgreSQL ai modelli LLM e di embedding disponibili su Amazon Bedrock, AWS mette a disposizione tutti gli ingredienti necessari per implementare una soluzione di Retrieval-Augmented Generation.

A fare la differenza, però, è la ricetta: dalla preparazione e suddivisione dei contenuti all'arricchimento dei metadati, fino alle strategie di retrieval e alla costruzione del contesto fornito al modello.

In questo talk analizzeremo l'implementazione di uno scenario reale, concentrandoci sulle scelte tecniche, sui compromessi emersi durante lo sviluppo e sul loro impatto sulla qualità delle risposte.

Senza tralasciare il tema dei costi, perché a fine mese arriva la fattura di AWS.

Bio

Andrea Saltarello è CEO di Improove e professore di Data Science e AI presso la Graduate School of Management del Politecnico di Milano. È fondatore e organizzatore di Cloud Day, conferenza italiana dedicata al cloud computing, e di AI Conf, evento di riferimento per professionisti, aziende e decision maker che lavorano con l'intelligenza artificiale. Da oltre vent'anni si occupa di architetture software, sistemi distribuiti e adozione delle tecnologie emergenti in ambito enterprise. Attraverso Improove sviluppa iniziative di formazione e confronto che mettono in relazione community tecniche, aziende e protagonisti dell'ecosistema cloud e AI.

Abstract

An open unconference session. Topics are proposed and chosen by attendees.