Spring AI Workshop for AWS Bedrock

Introduction

This is an introductory workshop for Spring AI with AWS Bedrock LLMs. Unlike vendors like Azure and Google, AWS has chosen to host multiple models rather than throw in entirely on their own model (Titan). The model I selected for this workshop is Anthropic3 for chat, and AWS's own Titan for embeddings.

NOTE: Many users will have experience doing AI workshops using python tools like langchain and streamlit. While that provides a nice GUI for testing, Spring is designed for enterprise applications and is almost always part of a larger application. Because Spring in general is a tool for developers, the entire workshop will be delivered through the command line terminal. While it is definitely possible to run command by posting commands such as http://localhost:8080/ai/simple in a browser, it is not really an optimal experience.

For best experience, we encourage user to deploy the .devcontainer, which has all necessary tools such as java, maven, http, and curl installed. Launching a CodeSpace from GitHub will open the devcontainer and bring you into the VsCode tool. In this case, we recommend running ./mvnw clean installbefore starting the first exercise to download all dependencies and prep the application. Also, it is not necessary to run ./mvnw spring-boot:run for every exercise if the app is still running.

Of course, users are welcome to download this repo and run it locally using their own tools as well. We also encourage users to visit Spring Initializer to recreate the code on their own.

Prerequisites

Before you begin make sure to set the following environment variables.

export AWS_ACCESS_KEY_ID=<INSERT ID HERE>
export AWS_SECRET_ACCESS_KEY=<INSERT KEY HERE>

NOTE: the Spring AI docs will tell you to enter these values in the /src/main/resources/application.properties file.
I found I had issues with the utility used in-house to access AWS, so I leave them out of application.properties opting for setting environmental variables in my console. That said, you must set the region in the application.properties file.

Enable Bedrock AI Model Access

The configuration assumes you have already enabled the required models inside your AWS Bedrock account. You can enable model access at the model access tab on the bottom-left of the main page. e.g. here. This workshop requires access to Anthropic for chat feature and Titan only for embeddings. At the time of this writing Anthropic does require a use case justification.

For the workshop, the configuration in application.properties should contain the following

spring.ai.bedrock.aws.region=<INSERT AWS REGION>

spring.ai.bedrock.anthropic3.chat.enabled=true

spring.ai.bedrock.titan.embedding.enabled=true
spring.ai.bedrock.titan.embedding.model=amazon.titan-embed-text-v1
spring.ai.bedrock.titan.embedding.input-type=TEXT

Workshop Overview

The workshop consists of six examples, each with a dedicated README file. Note that there is a description of how Spring, and the goal is to instruct user on how SpringAI functions. The code is essentially the same with every AI model.

All six workshop examples are organized into individual Java packages within this project. In each package, you'll find a Spring @RestController class that serves as the entry point for showcasing the discussed functionality.

To interact with the @RestController, you will be using the http utility as a user-friendly alternative to curl.

Detailed instructions and exercises for each example can be found in their respective README files:

• 1-README-tell-me-a-joke.md
• 2-README-prompt-templating.md
• 3-README-prompt-roles.md
• 4-README-output-parser.md
• 5-README-stuff-prompt.md
• 6-README-retrieval-augmented-generation.md

These guides will walk you through the workshop exercises.

NOTE: before running the workshop, please visit what is Spring to get some background on the Spring Framework and understand why Development teams love to use Spring AI.