Highlights of the AWS Cloud Day Zurich 2023

This article is also available in a German version.

In this blog, we present a summary of the AWS Cloud Day Zurich 2023. We provide an overview of the various activities of the event and summarize some presentations from the areas of generative artificial intelligence (generative AI), partner and customer sessions, builder sessions, and security.

The AWS team at AWS Cloud Day Zurich on the main stage

Facts & Figures

After the AWS Swiss Cloud Day and the start of the AWS Europe (Zurich) Region in 2022, it was that time again: AWS Cloud Day Zurich took place at the end of September 2023. We welcomed over 2,000 participants from various industries and professions in Halle 550 in Zurich Oerlikon. In more than 50 technical sessions, current topics were covered: From generative AI to serverless technologies, modern data architectures, SAP on AWS, security & governance, infrastructure as code and smart manufacturing.

There was also a Partner Expo with 27 booths, an AWS World with interactive demos, and a separate stage and exhibition area for AWS community & startups. With the Bundesliga Free Kick Challenge, AWS DeepRacer or the AWS BuilderCards Lounge, visitors were able to interact with technologies such as IoT, Machine Learning and Serverless in a playful way. 455 Serverlesspresso coffees were served and 225 laps were driven on AWS DeepRacer.

Archana Vemulapalli (Director of Product Management and Global Strategy for Data, Analytics and AI & ML, AWS) on the main stage and view of the audience during the keynote

Generative AI

Generative AI was a focus at AWS Cloud Day Zurich. “There’s no AI without the cloud” said AWS CEO Adam Selipsky recently. Solutions and approaches were shown in various presentations, while experts from AWS and AWS Partner Network (APN) partners were available for in-depth discussions at the booths.

The presentation “Augment your business with generative AI: myths, reality, and practice” started with an overview of the basics of generative AI. Generative AI is a technology using so-called Large Language Models (LLM), which is able to create various content, such as text, images, audio or code. LLMs typically do not (yet) have a specialization, but are general in nature. For this reason, they are often referred to as Foundation Models (FMs). To become context-aware, domain-specific data is required. For example: In order for a Foundation Model to build up contextual knowledge in the area of insurances, it must be enriched with data from the insurance sector. This data is often confidential (e.g. data containing company secrets), which is why the security and confidentiality of model and interaction data is particularly important. AWS services such as Amazon SageMaker or Amazon Bedrock make this task easier through various features such as encryption, access management, or logging and monitoring.

The presentation “Architecting AI: A tour of generative AI design patterns” presented the patterns used to integrate a generative AI application into a solution:

1. Using Amazon Bedrock, if you want easy handling
2. Using Amazon SageMaker Jumpstart if flexibility for adjustments is important
3. By running your own on Amazon Elastic Compute Cloud (Amazon EC2) or on containers running on services such as Amazon Elastic Container Service (Amazon ECS) to fully control integration

There were lots of insights and detailed information about specific use-cases for generative AI. These included creating content securely, using knowledge bases for models to deliver better results, and agents orchestrating complex processes and interactions with other systems.

The presentation “Architecting AI: Building generative AI applications” gave practical tips and instructions on how generative AI applications can be developed and what needs to be considered. This also includes avoiding toxicity in inputs and outputs and reducing model hallucinations (false statements). The presentation showed how a filter is created that can detect and block toxic inputs or outputs, using Amazon Comprehend and Amazon Rekognition. A specific example with demo code can also be found in the blog post “Safe image generation and diffusion models with Amazon AI content moderation services”. Retrieval-augmented Generation (RAG) or Reasoning + Acting (ReAct) can be used as measures to reduce model hallucinations.

APN Partner and AWS customer sessions

Partners and customers shared their experiences with AWS technologies in various talks.

In “Generative AI on AWS with Hugging Face”, AWS and Hugging Face demonstrated how easy it is to train and use generative AI models on AWS with Amazon SageMaker. Various Foundational Models from Hugging Face and other providers can be used as a basis for this. AWS offers highly performant and cost efficient instance types specifically built for AI workloads, using AWS Trainium and AWS Inferentia2 accelerators.

In “How to use AWS Outposts in Railway Operations”, AWS and the Swiss Federal Railways SBB showed how AWS services can be used within a customer data center using AWS Outposts. In addition to the customer’s requirements, the presentation also discussed how such a solution can be automated using Infrastructure-as-Code (IaC) with AWS CDK, ensuring a clear governance. Using AWS Outposts not only provides SBB with minimal latency, but also a consistent technology setup for on-premise applications. Thanks to the AWS Nitro system, customers also benefit from the usual high security standards of AWS services.

In the presentation “How Oerlikon uses Terraform for secure, automated governance”, Oerlikon and AWS showed how an IaC approach based on Terraform can be used to achieve a balance between development speed and security. As a result, Oerlikon was able to reduce the time it took their staff to create a new AWS account from weeks to hours. At the same time, both central governance and agility within the teams remain ensured.

In the presentation “How Ultumus (SIX Group) leverages AWS to become the bedrock of the Global ETF Industry”, Ultumus and AWS showed how Ultumus manages more than 3 million indices and 10,000 exchange-traded funds (ETFs) on AWS. Using serverless services such as AWS Lambda, Amazon DynamoDB, or Amazon Simple Storage Service (S3), Ultumus has built a highly available, automatically scalable and secure architecture. This means that Ultumus only pays for what is actually used and does not have to worry about operations, scaling, etc. itself.

Builder Sessions

The AWS Cloud Day Zurich is a great opportunity for developers to connect with experts and learn first-hand about new techniques and best practices. There were various sessions that addressed these topics.

One of them was of “Thinking asynchronously: Application integration patterns for microservices”. While synchronous processes and strong coupling were the norm in traditional architectures, asynchronous patterns are often better suited, particularly for highly scalable applications. This decoupling enables horizontal scaling, increases resilience and gives development teams more autonomy. Both choreography and orchestration can be used for processes, or a combination of both. AWS offers a range of services for event delivery in asynchronous architectures, such as Amazon EventBridge, Amazon Simple Queue Service (Amazon SQS), or AWS Step Functions.

The presentation “Define your cloud native disaster recovery strategy with AWS” showed how the global AWS infrastructure makes it easy for customers to prepare for failure scenarios. Using multiple Availability Zones provides, among other things, protection from natural events such as power outages, lightning strikes, earthquakes and more. For the unlikely event of an entire region failure, customers have the option to replicate to an additional region. Business continuity (Recovery Time Objective RTO, Recovery Point Objective RPO) and regulatory requirements as well as costs can influence the choice of the optimal approach. Options range from backups to an active-active setup and various approaches in-between.

Once a strategy has been defined and implemented, it must be thoroughly tested. This ensures that the measures actually work in the event of an error. The chaos engineering approach, which was examined in more detail in “Resilient and Well-Architected Apps with Chaos Engineering”, is suitable for this purpose. This discipline is about testing the resilience of an infrastructure by consciously introducing errors. AWS Fault Injection Simulator provides the tool for this. The service allows you to create experiments and thus create temporary errors in an application. The results of these experiments can then be used to make improvements. In order to avoid a negative impact on end customers, it is recommended to carry out these experiments in test environments first. If these are successful, they can also be carried out continuously in the production environment. Insights from production are real and offer maximum insight into the real resilience of an architecture. Amazon.com, for example, carries out such experiments on a daily basis, as explained in the blog “Any Day Can Be Prime Day: How Amazon.com Search Uses Chaos Engineering to Handle Over 84K Requests Per Second”.

Security

In line with our principle of “Security is our top priority,” there were also presentations on security and an according expert booth at this year’s AWS Cloud Day Zurich.

“Automating Compliance, Auditing & Evidence Collection on AWS” showed how cloud-native services can be used to facilitate compliance processes. Automation is the key to better scaling processes and making them reliably repeatable. AWS Config is used to continuously audit and evaluate compliance of resource configurations with an organization’s policies. AWS Audit Manager offers functionality to create links to controls catalogs (e.g. Payment Card Industry Data Security Standard, PCI DSS). Audit teams use these services to get a consolidated overview of all resources and their compliance status, while having the ability to dive deep and investigate.

“Building modern and secure applications on AWS” showed how AWS services can be used to protect modern applications. With little effort, developers can leverage a web application firewall (AWS WAF) to protect web applications and AWS X-Ray to trace and analyse distributed applications. With Amazon CloudWatch Synthetics, they can use “canaries” to constantly monitor availability and latency of applications. In addition, Amazon CloudWatch RUM provides insights into the user experience, by providing measurements like page load time or client-side errors. The measured data provides a foundation for increasing the availability, resilience and usability of a modern web application.

Summary

The AWS Cloud Day Zurich 2023 brought AWS, partners, experts and customers together for one day. It was a day full of interesting conversations, educational talks and demonstrations. Come by and visit us at the next AWS Event!