Skip to content

Latest commit

 

History

History
217 lines (151 loc) · 14.8 KB

readme.md

File metadata and controls

217 lines (151 loc) · 14.8 KB
Raw

GenAI Spring PetClinic Sample Application build with LangChain4j Build Status

Open in Gitpod Open in GitHub Codespaces

Understanding the Spring Petclinic LangChain4j application

A chatbot using Generative AI has been added to the famous Spring Petclinic application. This version uses the LangChain4j project and currently supports OpenAI or Azure's OpenAI as the LLM provider. This is a fork from the spring-petclinic-ai based on Spring AI.

This sample demonstrates how to easily integrate AI/LLM capabilities into a Java application using LangChain4j. This can be achieved thanks to:

  • A unified abstraction layer designed to decouple your code from specific implementations like LLM or embedding providers, enabling easy component swapping. Only the application.properties file references LLM providers such as OpenAI or Azure OpenAI.
  • Memory offers context to the LLM for both your current and previous conversations, with support for multiple users. Refer to the use of the MessageWindowChatMemory class in AssistantConfiguration and the @MemoryId annotation in the Assistant interface.
  • AI Services enables declarative definitions of complex AI behaviors through a straightforward Java API. See the use of the @AiService annotation in the Assistant interface.
  • System prompts play a vital role in LLMs as they shape how models interpret and respond to user queries. Look at the @SystemMessage annotation usage in the Assistant interface.
  • Streaming response token-by-token when using the TokenStream return type and Spring Server-Sent Events supports. Take a look at the AssistantController REST controller
  • Function calling or Tools allows the LLM to call, when necessary, one or more java methods. The AssistantTool component declares functions using the @Tool annotation from LangChain4j.
  • Structured outputs allow LLM responses to be received in a specified format as Java POJOs. AssistantTool uses Java records as the LLM/ input/output data structure.
  • Retrieval-Augmented Generation (RAG) enables an LLM to incorporate and respond based on specific data—such as data from the petclinic database—by ingesting and referencing it during interactions. The AssistantConfiguration declares the EmbeddingModel, InMemoryEmbeddingStore and EmbeddingStoreContentRetrieverbeans while the EmbeddingStoreInit class handles vets data ingestion at startup. The VetQueryRouter demonstrates how to conditionally skip retrieval, with decision-making driven by an LLM.

Spring Petclinic integrates a Chatbot that allows you to interact with the application in a natural language. Here are some examples of what you could ask:

  1. Please list the owners that come to the clinic.
  2. How many veterinary cardiologists are there?
  3. Is there an owner named Betty? What's her lastname?
  4. Which owners have dogs?
  5. Add a dog for Betty. Its name is Moopsie. His birthday is on 2 October 2024.
  6. Add today's visit to Moopsie.

Screenshot of the chat dialog

Spring Petclinic currently supports OpenAI or Azure's OpenAI as the LLM provider. In order to start spring-petlinic-langchain4j perform the following steps:

  1. Decide which provider you want to use. By default, the langchain4j-open-ai-spring-boot-starter dependency is enabled. You can change it to langchain4j-azure-open-ai-spring-boot-starterin eitherpom.xml or in build.gradle, depending on your build tool of choice.
  2. Create an OpenAI API key or a Azure OpenAI resource in your Azure Portal. Refer to the OpenAI's quickstart or Azure's documentation for further information on how to obtain these. You only need to populate the provider you're using - either openai, or azure-openai. If you don't have your own OpenAI API key, don't worry! You can temporarily use the demo key, which OpenAI provides free of charge for demonstration purposes. This demo key has a quota, is limited to the gpt-4o-mini model, and is intended solely for demonstration use.
  3. Export your API keys and endpoint as environment variables:
    • either OpenAI:
    export OPENAI_API_KEY="your_api_key_here"
    • or Azure OpenAI:
    export AZURE_OPENAI_ENDPOINT="https://your_resource.openai.azure.com"
export AZURE_OPENAI_KEY="your_api_key_here"
  4. Follow the next section Run Petclinic locally

Run Petclinic locally

Spring Petclinic is a Spring Boot application built using Maven or Gradle. You can build a jar file and run it from the command line (it should work just as well with Java 17 or newer):

git clone https://github.com/spring-petclinic/spring-petclinic-langchain4j.git
cd spring-petclinic
./mvnw package
java -jar target/*.jar

You can then access the Petclinic at http://localhost:8080/.

Screenshot of the Find Owners menu

Or you can run it from Maven directly using the Spring Boot Maven plugin. If you do this, it will pick up changes that you make in the project immediately (changes to Java source files require a compile as well - most people use an IDE for this):

./mvnw spring-boot:run

NOTE: If you prefer to use Gradle, you can build the app using ./gradlew build and look for the jar file in build/libs.

Building a Container

There is no Dockerfile in this project. You can build a container image (if you have a docker daemon) using the Spring Boot build plugin:

./mvnw spring-boot:build-image

In case you find a bug/suggested improvement for Spring Petclinic

Our issue tracker is available here.

Database configuration

In its default configuration, Petclinic uses an in-memory database (H2) which gets populated at startup with data. The h2 console is exposed at http://localhost:8080/h2-console, and it is possible to inspect the content of the database using the jdbc:h2:mem:<uuid> URL. The UUID is printed at startup to the console.

A similar setup is provided for MySQL and PostgreSQL if a persistent database configuration is needed. Note that whenever the database type changes, the app needs to run with a different profile: spring.profiles.active=mysql for MySQL or spring.profiles.active=postgres for PostgreSQL. See the Spring Boot documentation for more detail on how to set the active profile.

You can start MySQL or PostgreSQL locally with whatever installer works for your OS or use docker:

docker run -e MYSQL_USER=petclinic -e MYSQL_PASSWORD=petclinic -e MYSQL_ROOT_PASSWORD=root -e MYSQL_DATABASE=petclinic -p 3306:3306 mysql:8.4

or

docker run -e POSTGRES_USER=petclinic -e POSTGRES_PASSWORD=petclinic -e POSTGRES_DB=petclinic -p 5432:5432 postgres:16.3

Further documentation is provided for MySQL and PostgreSQL.

Instead of vanilla docker you can also use the provided docker-compose.yml file to start the database containers. Each one has a profile just like the Spring profile:

docker-compose --profile mysql up

or

docker-compose --profile postgres up

Test Applications

At development time we recommend you use the test applications set up as main() methods in PetClinicIntegrationTests (using the default H2 database and also adding Spring Boot Devtools), MySqlTestApplication and PostgresIntegrationTests. These are set up so that you can run the apps in your IDE to get fast feedback and also run the same classes as integration tests against the respective database. The MySql integration tests use Testcontainers to start the database in a Docker container, and the Postgres tests use Docker Compose to do the same thing.

Compiling the CSS

There is a petclinic.css in src/main/resources/static/resources/css. It was generated from the petclinic.scss source, combined with the Bootstrap library. If you make changes to the scss, or upgrade Bootstrap, you will need to re-compile the CSS resources using the Maven profile "css", i.e. ./mvnw package -P css. There is no build profile for Gradle to compile the CSS.

Working with Petclinic in your IDE

Prerequisites

The following items should be installed in your system:

Steps

  1. On the command line run:

    git clone https://github.com/spring-petclinic/spring-petclinic-langchain4j.git

  2. Inside Eclipse or STS:

    Open the project via File -> Import -> Maven -> Existing Maven project, then select the root directory of the cloned repo.

    Then either build on the command line ./mvnw generate-resources or use the Eclipse launcher (right-click on project and Run As -> Maven install) to generate the CSS. Run the application's main method by right-clicking on it and choosing Run As -> Java Application.

  3. Inside IntelliJ IDEA:

    In the main menu, choose File -> Open and select the Petclinic pom.xml. Click on the Open button.

    • CSS files are generated from the Maven build. You can build them on the command line ./mvnw generate-resources or right-click on the spring-petclinic project then Maven -> Generates sources and Update Folders.

    • A run configuration named PetClinicApplication should have been created for you if you're using a recent Ultimate version. Otherwise, run the application by right-clicking on the PetClinicApplication main class and choosing Run 'PetClinicApplication'.

  4. Navigate to the Petclinic

    Visit http://localhost:8080 in your browser.

Looking for something in particular?

Spring Boot Configuration Class or Java property files
The Main Class PetClinicApplication
Properties Files application.properties
Caching CacheConfiguration

Interesting Spring Petclinic branches and forks

The Spring Petclinic "main" branch in the spring-projects GitHub org is the "canonical" implementation based on Spring Boot and Thymeleaf. There are quite a few forks in the GitHub org spring-petclinic. If you are interested in using a different technology stack to implement the Pet Clinic, please join the community there.

Interaction with other open-source projects

One of the best parts about working on the Spring Petclinic application is that we have the opportunity to work in direct contact with many Open Source projects. We found bugs/suggested improvements on various topics such as Spring, Spring Data, Bean Validation and even Eclipse! In many cases, they've been fixed/implemented in just a few days. Here is a list of them:

Name Issue
Spring JDBC: simplify usage of NamedParameterJdbcTemplate SPR-10256 and SPR-10257
Bean Validation / Hibernate Validator: simplify Maven dependencies and backward compatibility HV-790 and HV-792
Spring Data: provide more flexibility when working with JPQL queries DATAJPA-292

Contributing

The issue tracker is the preferred channel for bug reports, feature requests and submitting pull requests.

For pull requests, editor preferences are available in the editor config for easy use in common text editors. Read more and download plugins at https://editorconfig.org. If you have not previously done so, please fill out and submit the Contributor License Agreement.

License

The Spring PetClinic sample application is released under version 2.0 of the Apache License.