Skip to content

Latest commit

 

History

History
80 lines (68 loc) · 4.76 KB

README.md

File metadata and controls

80 lines (68 loc) · 4.76 KB
Raw

How a Picture of Car Damage Can File Your Insurance Claim

1. Image Analysis for Claim Data Extraction

This initial step uses GPT-4 Turbo with Vision to analyze images submitted for insurance claims. It extracts information such as the type of car damage, the car's make and model, and the license plate number. This data is foundational for processing the claim.

2. Insurance Coverage Determination

In this phase, the semantic kernel is tasked with verifying whether the specific car insurance policy covers the identified damage. Utilizing the Stepwise Planner, we can direct the kernel to utilize the InsurancePolicyPlugin. The goal is to find the relevant insurance policy and confirm coverage for the particular incident. This crucial step directly addresses whether the insurance policy covers the reported incident.

String ENDPOINT = System.getenv("AZURE_OPEN_AI_ENDPOINT");
String API_KEY = System.getenv("AZURE_OPEN_AI_KEY");
String DEPLOYMENT_NAME = System.getenv("GPT4_DEPLOYMENT_NAME");

OpenAIAsyncClient client = new OpenAIClientBuilder()
.endpoint(ENDPOINT)
.credential(new AzureKeyCredential(API_KEY))
.buildAsyncClient();

Kernel kernel = SKBuilders.kernel()
.withDefaultAIService(SKBuilders.chatCompletion()
.withOpenAIClient(client)
.withModelId(DEPLOYMENT_NAME)
.build())
.build();

kernel.importSkill(new InsurancePolicyPlugin(), "InsurancePolicyPlugin");
var goal = "Your goal is by using the InsurancePolicyPlugin to 1. find out which policy relates to the car accident report and 2. if the policy covers parking damage. Provide the policy information related to the car (if there is any) and inform the user if the damage is covered or not. Use an emoji in your response. \nInsurance Claim Information: "
+ responseText;
StepwisePlanner planner = new DefaultStepwisePlanner(kernel, null, null);
var plan = planner.createPlan(goal);
SKContext result = plan.invokeAsync().block();
return result.getResult();

3. Custom Plugin Integration for Policy Verification

The InsurancePolicyPlugin is a custom plugin created for the kernel. It connects with the existing (mocked) policy system to fetch insurance policies. This plugin exemplifies a typical developer task when building AI apps - creating plugins that extend the kernel's capabilities by integrating with existing systems.

public class InsurancePolicyPlugin {
    private final HttpClient httpClient;
    private final String baseUrl = "http://localhost:8080/api/insurance-policies";
      // Constructor initializes the HttpClient
      public InsurancePolicyPlugin() {
          this.httpClient = HttpClient.createDefault();
      }
    
      // Function to fetch insurance policies
      @DefineSKFunction(description = "Fetch the list of insurance policies", name = "GetPolicies")
      public Mono<String> getPoliciesAsync(SKContext context) {
          HttpRequest request = new HttpRequest(HttpMethod.GET, baseUrl);
          return httpClient.send(request).flatMap(response -> response.getBodyAsString());
      }
}

Preview

Prerequisites

To deploy this application, please ensure you have the following dependencies installed:

Setup Procedure

  1. Clone the repository to your local machine.
  2. Navigate to the project's root directory.
  3. Run mvn clean install to build the project using Maven.
  4. Before starting the application, set the endpoint and key:
    • export AZURE_OPEN_AI_ENDPOINT=https://<instance-name>.openai.azure.com/
    • export AZURE_OPEN_AI_KEY=<your-key>
    • export GPT4_VISION_DEPLOYMENT_NAME=<your-gpt4-vision-deployment-name>
    • export GPT4_DEPLOYMENT_NAME=<your-gpt4-deployment-name>
  5. Start the application with java -jar target/appname.jar and access it via http://localhost:8080.

Future Improvements

  • Implement GPT-4 Vision analysis directly within Semantic Kernel (once supported).