Semantic Kernel Quick-Start for Web API

Semantic Kernel Quick-Start for Blazor is a CRUD Microservice solution that demonstrates the most basic use cases of the Microsoft Semantic Kernel in a Clean Architecture microservice. Chat Completion, Text Generation, Text to Audio, Text to Image.

This microservice allows you to persist the following Azure Open AI services to SQL Server, so you can replay messages and maintain history of your interaction with AI.

Semantic Kernel is an SDK that integrates Large Language Models (LLMs) like OpenAI, Azure OpenAI, and Hugging Face with conventional programming languages like C#, Python, and Java. Semantic Kernel allows developers to define plugins that can be chained together in just a few lines of code.

Getting-Started

To get started, follow the steps below:

1. Install Prerequisites (Visual Studio, dotnet, ef and Semantic Kernel Quick-Start from Visual Studio marketplace)
2. In Visual Studio, File -> New Project -> Semantic Kernel Quick-start
3. Add your Open AI or Azure Open AI key to configuration (via dotnet user-secrets set command)
4. Create your SQL Server database (via dotnet ef command)
5. Run tests (via dotnet test command or Test Explorer)

A bit about Semantic Kernel

Introduction to Semantic Kernel

Getting Started with Semantic Kernel

This microservice supports:

• Chat Completions: Generate responses based on user input, making it useful for chatbots and virtual assistants.
• Text to Speech: Convert text into natural-sounding speech, enhancing user experiences.
• Whisper (Text to Speech): Convert spoken language into text, useful for transcription and voice recognition.
• Image to Text: Generate descriptions of an image.
• Text to Image: Create an image based on a description prompt of the desired imagery.

Upcoming relases will support more Semantic Kernel and Azure Open AI functionality such as:

• Embeddings: Create vector representations of text, which can be used for semantic search and similarity matching.
• Function Calling: Integrate custom functions into your AI models, allowing the model to call external APIs or perform specific tasks based on the context of the conversation.
• Content Filtering: Automatically filter out inappropriate or harmful content from generated responses.
• Fine-Tuning: Train models on your specific data to better align with your use cases and improve performance.
• Assistants: Create and manage virtual assistants that can handle complex tasks and interactions.
• Semantic Search: Perform searches based on the meaning of the text rather than just keywords, improving search relevance.

Install Prerequisites

You will need the following tools:

Visual Studio

Visual Studio Workload IDs

winget install --id Microsoft.VisualStudio.2022.Community --override "--quiet --add Microsoft.Visualstudio.Workload.Azure --add Microsoft.VisualStudio.Workload.Data --add Microsoft.VisualStudio.Workload.ManagedDesktop --add Microsoft.VisualStudio.Workload.NetWeb"

Or VS Code (code .)

winget install Microsoft.VisualStudioCode --override '/SILENT /mergetasks="!runcode,addcontextmenufiles,addcontextmenufolders"'

.NET SDK

winget install Microsoft.DotNet.SDK.8 --silent

dotnet ef cli

Install

dotnet tool install --global dotnet-ef

Update

dotnet tool update --global dotnet-ef

SQL Server

Visual Studio installs SQL Express. If you want full-featured SQL Server, install the SQL Server Developer Edition or above.

SQL Server Developer Edition or above

Configure API Key and Connection String

Follow these steps to get your development environment set up:

ASPNETCORE_ENVIRONMENT set to "Local" in launchsettings.json

1. This project uses the following ASPNETCORE_ENVIRONMENT to set configuration profile

• Debugging uses Properties/launchSettings.json
• launchSettings.json is set to Local, which relies on appsettings.Local.json

2. As a standard practice, set ASPNETCORE_ENVIRONMENT entry in your Enviornment Variables and restart Visual Studio

    Set-Item -Path Env:ASPNETCORE_ENVIRONMENT -Value "Development"
    Get-Childitem env:
   

Setup Azure Open AI or Open AI configuration

Important: Do this for both Presentation.WebAPI and Specs.Infrastructure

Azure Open AI

cd src/Presentation/WebAPI
dotnet user-secrets init
dotnet user-secrets set "AzureOpenAI:ChatDeploymentName" "gpt-4"
dotnet user-secrets set "AzureOpenAI:Endpoint" "https://YOUR_ENDPOINT.openai.azure.com/"
dotnet user-secrets set "AzureOpenAI:ApiKey" "YOUR_API_KEY"

Alternately you can set in Environment variables

AzureOpenAI__ChatDeploymentName
AzureOpenAI__Endpoint
AzureOpenAI__ApiKey

Open AI

Set API Key in both Presention/WebAPI and Tests/Specs.Integration projects

cd src/Presentation/WebAPI
dotnet user-secrets init
dotnet user-secrets set "OpenAI:ApiKey" "YOUR_API_KEY"
cd src/Tests/Specs.Integration
dotnet user-secrets init
dotnet user-secrets set "OpenAI:ApiKey" "YOUR_API_KEY"

Alternately you can set in Environment variables

OpenAI__ChatModelId	
OpenAI__ApiKey

Setup your SQL Server connection string

dotnet user-secrets init
dotnet user-secrets set "ConnectionStrings:DefaultConnection" "YOUR_SQL_CONNECTION_STRING"

Create SQL Server Database

dotnet ef migrate steps

1. Open Windows Terminal in Powershell or Cmd mode

2. cd to root of repository

3. Deploy entities and configurations

    dotnet ef database update --project .\src\Infrastructure\SqlServer\Infrastructure.SqlServer.csproj --startup-project .\src\Presentation\WebApi\Presentation.WebApi.csproj --context SemanticKernelContext --connection "Data Source=(localdb)\MSSQLLocalDB;Initial Catalog=SemanticKernelMicroservice;Min Pool Size=3;MultipleActiveResultSets=True;Trusted_Connection=Yes;TrustServerCertificate=True;MultipleActiveResultSets=False;Encrypt=True;TrustServerCertificate=False;Connection Timeout=30"
   

Running the Application

Launch the backend

Right-click Presentation.WebApi and select Set as Default Project

dotnet run Presentation.WebApi.csproj

Open http://localhost:7777/swagger/index.html

Open Microsoft Edge or modern browser Navigate to: http://localhost:7777/swagger/index.html in your browser to the Swagger API Interface

DevOps Configuration for Azure IaC and CI/CD

GitHub Actions (.github folder)

The GitHub action will automatically run upon commit to a repo. The triggers are set based on changes (PRs/Merges) to the main branch.

Azure Federation to GitHub Actions

gtc-rg-semantickernel-infrastructure.yml will deploy all necessary resources to Azure. To enable this functionality, two service principles are required: App Registration service principle (used for az login command) and a Enterprise Application service principle (allows GitHub to authenticate to Azure).

Git Hub Environment Secret setup and Azure IAM privileges:

Note: The AZURE_SECRETS method uses: az ad sp create-for-rbac --name "COMPANY-SUB_OR_PRODUCTLINE-github-001" --role contributor --scopes /subscriptions/SUBSCRIPTION_ID --json-auth

New-AzureGitHubFederation.ps1

Install-Module Az #-Force #Force will update the module if it is already installed

Connect-AzAccount -SubscriptionId $SubscriptionId -UseDeviceAuthentication

# Create a new Azure AD App Registration application and service principal
$existingAppRegistration = Get-AzADApplication -Filter "displayName eq '$PrincipalName'"
if (-not $existingAppRegistration) {
    New-AzADApplication -DisplayName $PrincipalName
}
$clientId = (Get-AzADApplication -DisplayName $PrincipalName).AppId

New-AzADServicePrincipal -ApplicationId $clientId
$objectId = (Get-AzADServicePrincipal -DisplayName $PrincipalName).Id

New-AzRoleAssignment -ObjectId $objectId -RoleDefinitionName Contributor -Scope "/subscriptions/$SubscriptionId"
$clientId = (Get-AzADApplication -DisplayName $PrincipalName).AppId
$tenantId = (Get-AzContext).Subscription.TenantId

# Create new App Registration Federated Credentials for the GitHub operations
$subjectRepo = $subjectRepo = "repo:" + $Organization + "/" + $Repository + ":environment:" + $Environment

New-AzADAppFederatedCredential -ApplicationObjectId $objectId -Audience api://AzureADTokenExchange -Issuer 'https://token.actions.githubusercontent.com' -Name "$PrincipalName-repo" -Subject "$subjectRepo"
$subjectRepoMain = "repo:" + $Organization + "/" + $Repository + ":ref:refs/heads/main"

New-AzADAppFederatedCredential -ApplicationObjectId $objectId -Audience api://AzureADTokenExchange -Issuer 'https://token.actions.githubusercontent.com' -Name "$PrincipalName-main" -Subject "$subjectRepoMain"
$subjectRepoPR = "repo:" + $Organization + "/" + $Repository + ":pull_request"

New-AzADAppFederatedCredential -ApplicationObjectId $objectId -Audience api://AzureADTokenExchange -Issuer 'https://token.actions.githubusercontent.com' -Name "$PrincipalName-PR" -Subject "$subjectRepoPR"

In GitHub repo environment: Add the az login secrets:

• AZURE_CLIENT_ID
• AZURE_TENANT_ID
• AZURE_SUBSCRIPTION_ID

Azure DevOps Pipelines (.azure-devops folder)

Azure DevOps pipelines require an Azure Service Connection to authenticate and deploy resources to Azure.

Entity Framework vs. Semantic Kernel Memory

This example uses Entity Framework (EF) to store messages and responses for Semantic Kernel, and does not rely on SK Memory (SM). EF and SM serve different purposes. If you need natural language querying and efficient indexing, Semantic Kernel Memory is a great fit. If you’re building a standard application with a relational database, Entity Framework is more appropriate.

The key differences between Entity Framework (EF) and Semantic Kernel memory:

Purpose and Functionality

• Entity Framework (EF): EF is an Object-Relational Mapping (ORM) framework for .NET applications. It allows you to map database tables to C# classes and provides an abstraction layer for database operations. EF focuses on CRUD (Create, Read, Update, Delete) operations and data modeling.
• Semantic Kernel Memory: Semantic Kernel Memory (SM) is part of the Semantic Kernel project. It’s a library for C#, Python, and Java that wraps direct calls to databases and supports vector search. SM is designed for long-term memory and efficient indexing of datasets. It’s particularly useful for natural language querying and retrieval augmented generation (RAG).

Data Storage and Retrieval

• EF: EF stores data in relational databases (e.g., SQL Server, MySQL, PostgreSQL). It uses SQL queries to retrieve data.
• SM: SM can use various storage mechanisms, including vector databases. It supports vector search, which allows efficient similarity-based retrieval. SM is well-suited for handling large volumes of data and complex queries.

Querying

• EF: EF queries are typically written in LINQ (Language Integrated Query) or SQL. You express queries in terms of C# objects and properties.
• SM: SM supports natural language querying. You can search for information using text-based queries, making it more user-friendly for applications like chatbots.

Integration with Chat Systems

• EF: EF doesn’t directly integrate with chat systems. It’s primarily used for data persistence.
• SM: SM integrates seamlessly with chat systems like ChatGPT, Copilot, and Semantic Kernel. It enhances data-driven features in AI applications.

Scalability and Performance

• EF: EF is suitable for small to medium-sized applications. It may not perform optimally with very large datasets.
• SM: SM is designed for scalability. It can handle large volumes of data efficiently, making it suitable for memory-intensive applications.

Use Cases

• EF: Use EF for traditional CRUD operations, business logic, and data modeling.
• SM: Use SM for long-term memory, chatbots, question-answering systems, and information retrieval.

Contact

• GitHub Repo
• @goodtocode
• github.com/goodtocode

Technologies

• ASP.NET .Net
• Entity Framework Core
• MediatR
• AutoMapper
• Specflow
• FluentValidation
• FluentAssertions
• Moq

Semantic Kernel

• GitHub
• Getting Started Blog
• Understanding the Kernel
• Creating Plugins

Additional Technologies References

• AspNetCore.HealthChecks.UI
• Entity Framework Core
• FluentValidation.AspNetCore
• Microsoft.AspNetCore.App
• Microsoft.AspNetCore.Cors
• Swashbuckle.AspNetCore.SwaggerGen
• Swashbuckle.AspNetCore.SwaggerUI

Version History

This project is licensed with the MIT license.