Building Generative AI Applications with SageMaker Foundational Models

Discover the power of AWS SageMaker Jumpstart Notebooks in this hands-on workshop. Learn to deploy foundational models for text & image generation, integrate them with Lambda, and create a front-end application. Explore how to harness AI for building creative, scalable, and user-friendly solutions. No prior AI/ML experience required. You will be provided AWS Accounts

Lessons from the Trenches: Engineering Lessons from Everyday Startups

Engineering teams are often judged by the number of features they deploy, creating a feedback loop where more features equal a better team. However, true value goes beyond feature count. I'll share stories from startups where minor oversights led to significant repercussions, such as a missed service feature announcement costing tens of thousands of dollars, and outdated service architectures draining time and resources. Attendees will learn about refactoring and technical debt reduction by improving existing codebases and architectures periodically to prevent costly technical debt, and fostering a culture of learning and continuous improvement by empowering teams to share lessons, experiment with new technologies, and refine practices. Join me to explore how these aspects can lead to more resilient, efficient, and successful engineering teams, even outside unicorn startups.

It’s not the network, until it is: Mastering native tools

Facing “the network is down” panic? This dev chat dives into troubleshooting a complex AWS environment where your application can’t connect over AWS Direct Connect. Cut through the maze of routing, firewalls, and transit gateways using Amazon VPC Network Access Analyzer, VPC Reachability Analyzer, and VPC Flow Logs. Learn to quickly pinpoint and resolve issues in a hybrid network setup. Ideal for those with a basic understanding of AWS routing (VPCs, subnets, and transit gateways), this dev chat equips you with practical skills to efficiently troubleshoot without getting lost in configurations.

Full-stack AI with AWS Amplify

Building generative AI functionality is more than just calling an LLM, you need to be able to call it from your frontend application, authenticate and authorize requests, and hook everything up to a database to save conversations and use data for RAG. In this talk you'll learn how to build an end-to-end application with all of the above in under 30 minutes using AWS Amplify's new generative AI functionality.

Rethinking Serverless

There has been lots of discourse across the internet about Serverless in recent years. Some unsolicted advice its true while other pieces of advice have some elements of truth to them depending on the circumstances, and some is just plain incorrect. In this talk, we’ll cover some of the most common topics surrounding Serverless discussed on social media along with Danielle’s opinion based off of her real live experience of using Serverless in production workloads.

AI/ML Development on AWS - What Model Should I Use?

We represent data ops, data engineering, and data science on a two-pizza data team at SmugMug. Within our company, a unique contribution by our team is an AWS AI/ML sandbox, based on SageMaker. We'd like to show you how we put that together, and how other teams at the company use it. Highlights are structured access to the data lake, policy-driven control of compute sizes, and GenAI-enhanced data catalogs so our colleagues quickly understand the data they're working with. We will demo how to use the sandbox to train, evaluate, and start the process of deploying models. With the rise of so many great third-party AI models, a critical use case has become rapid internal benchmarking of new models against our own data, so we can decide which one is the best fit for our application.

Visualize and design your serverless applications

Discover how to start building serverless web applications that can solve common problems. In this session, learn how to get started with AWS Application Composer, a low-code, visual interface for designing and building serverless applications. Application Composer helps builders understand their application architecture, collaborate, and manage application configuration. Also, learn how to prototype a serverless generative AI application leveraging Amazon Bedrock from concept to a fully featured application using Application Composer and Amazon Q Developer in your IDE.

Optimizing GenAI Workloads on AWS

As GenAI models grow more complex, successfully optimizing their accuracy, cost efficiency, and deployment velocity on AWS becomes both increasingly critical and challenging. This session will provide an end-to-end guide for GenAI teams to maximize their workload efficiency on AWS. We’ll provide an overview of the full GenAI development lifecycle then do a deep dive into progressive model optimization techniques—from prompt engineering to RAG and fine tuning—exploring how each incrementally improves accuracy while minimizing retraining compute requirements. Shifting gears, we’ll share AWS best practices for optimizing GenAI model packaging, deployment, and inference using containers and hardware acceleration. Monitoring and maintaining production GenAI workloads presents its own challenges, which we’ll address through observability, data drift detection, and model degradation monitoring techniques and tools on AWS. Attendees will walk away with a clear framework for incrementally optimizing their GenAI workloads throughout the machine learning lifecycle with an eye toward maximizing performance while keeping costs in check.

Use Generative AI and Next.js with AWS Amplify to build a Fullstack Recipe Generator

Let’s dive into the world of Generative AI, Next.js, AWS Amplify, and Amazon Bedrock supercharged by Claude 3. In this guide, we’ll walk you through creating a recipe generator app where users can input a list of ingredients, and Claude 3 will generate delicious recipes based on their selection.

It is time for a new style of AWS Agencies

I recently opened an AWS agency to help clients with their AWS projects. I have a different approach than most agencies I have worked in. In this session, I would like to share my thoughts on how I think an AWS agency should be run and what I think is important when working with customers. View more or less new ideas like: - hiring from Meetups - self-organized Employers - direct revenue sharing with employees and some more, I will also share a customer success story working with this new approach.

Optimizing GenAI Workloads on AWS

As GenAI models grow more complex, successfully optimizing their accuracy, cost efficiency, and deployment velocity on AWS becomes both increasingly critical and challenging. This session will provide an end-to-end guide for GenAI teams to maximize their workload efficiency on AWS. We’ll provide an overview of the full GenAI development lifecycle then do a deep dive into progressive model optimization techniques—from prompt engineering to RAG and fine tuning—exploring how each incrementally improves accuracy while minimizing retraining compute requirements. Shifting gears, we’ll share AWS best practices for optimizing GenAI model packaging, deployment, and inference using containers and hardware acceleration. Monitoring and maintaining production GenAI workloads presents its own challenges, which we’ll address through observability, data drift detection, and model degradation monitoring techniques and tools on AWS. Attendees will walk away with a clear framework for incrementally optimizing their GenAI workloads throughout the machine learning lifecycle with an eye toward maximizing performance while keeping costs in check.

CI/CD: GitHub Actions to ECS

Peter will walk through how he has set up a CI/CD pipeline using GitHub Actions deploying to ECS. If you have ever been curious about setting up continuous deployment, this talk will give you the foundation to get started.

The Bedrock of RAG

Retrieval Augmented Generation (RAG) is one of the hottest topics for making data more accessible and valuable. And it's easy to get started with it on AWS, especially when using Bedrock as the foundation for the Large Language Model (LLM). This talk shows how to get started with RAG on AWS, using a playground to go from custom data to an interactive application in minutes. The second part of the demo then dives into more advanced topics around Bedrock and how to tune your RAG's generation part: * How does the chosen LLM influence quality and performance? * Can you tune the prompt? * Can the LLM also help with the retrieval step? At the end of this session, you will better understand of how to get started and how to get the most out of Bedrock with your AWS-powered RAG.

Simplifying AWS Services Access for pods with EKS Pod Identity

EKS Pod Identity is a new feature released by AWS in November 2023, which simplifies the configuration of IAM permissions for pods hosted on Amazon Elastic Kubernetes Service (EKS). We will compare EKS Pod Identity with previous methods, like IAM Role for Service Accounts (IRSA), highlighting the benefits of the new feature, such as eliminating the need for an OIDC Provider and simplifying the association of a Service Account with an IAM role. We will also explain how EKS Pod Identity uses IAM Session Tags to control the reuse of an IAM role for pods running on different clusters, namespaces, or using different Service Accounts.

Higher Order Abstraction & Tooling for Step Functions & Serverless

Serverless services like API Gateways, Step Functions, Lambdas, and EventBridge transform the Cloud into a MIMD Computer, but existing tools treat them as flat infrastructure. We need better abstractions and tools to manage serverless event-driven architectures through composition. At Informed, we process hundreds of thousands of loan documents daily using a 100% serverless architecture with 50 Step Functions, over 200 Lambdas, and 200+ EventBridge rules. Existing tools like SAM, SST, and Terraform are too verbose and indirect. We developed a YAML/JSON declarative toolset (soon to be open-sourced) to define and compose modular services into higher-order services, decoupling service definition from deployment. The talk includes a short demo.

Leverage ML Inference for Generative AI Models on AWS

LLM is becoming essential for many companies - either as a core product, an internal tool, or as a service for improving operations. One challenge when deploying the LLM to production is navigating through different hardware, service, and orchestration options. This presentation is focused on providing a comprehensive understanding of different LLM deployment options in AWS Cloud. We will explore different ways of deploying pretrained models to the AWS cloud from SageMaker Real-time inference to Elastic Container Service and AWS Lambda. We will also cover different ways of deploying ML infrastructure - from SageMaker JumpStart and SageMaker SDK to AWS Copilot and AWS SAM and options like Amazon Bedrock. There will be a live demo that shows how the deployment process would look like for the generative AI model using SageMaker JumpStart, ECS/Fargate + AWS Copilot and for AWS Lambda.