repository·main·Indexed 21 days ago
https://github.com/dotnet/ai-samplesA collection of .NET samples demonstrating the integration of AI capabilities (LLMs, SLMs, and embeddings) using unified abstractions and SDKs. It features Microsoft.Extensions.AI for provider-agnostic C# abstractions, Microsoft.Extensions.AI.Evaluation for response quality tooling, and practical implementations using OpenAI, Azure OpenAI, Ollama, and Semantic Kernel. Samples include text summarization, chat applications, function calling, image generation with DALL-E 3, and Semantic Kernel integration in web apps.
This repository provides samples demonstrating how to use the OpenAI reference implementation via the Microsoft.Extensions.AI.OpenAI NuGet package.
Available sample types include:
OpenAIExamples directory, these provide a set of samples for interacting with the OpenAI implementation.OpenAIWebAPI directory, these demonstrate a minimal Web API implementation using the same package.Microsoft.Extensions.AI.Ollama NuGet package. It includes both console applications and Web API implementations to show how to integrate local LLMs into .NET applications using the Microsoft.Extensions.AI abstractions.This project provides practical examples of how to use the Microsoft.Extensions.AI.Evaluation ecosystem to perform 'offline' evaluations. It demonstrates how to combine evaluators (Quality, Safety) with reporting concepts like response caching, result storage, and report generation.
Key capabilities demonstrated include:
dotnet test to run evaluations locally or in CI/CD pipelines.Microsoft.Extensions.AI.Evaluation.Reporting to store LLM responses in a cache. This reduces costs and improves performance by reusing responses for identical request parameters (model, endpoint, prompts, context).This sample demonstrates how to implement and manage chat history within a .NET AI application. It provides a practical implementation of maintaining context across multiple turns in a conversation.
For complete, step-by-step documentation, including implementation details and exercise instructions, refer to the full exercise guide.
Microsoft.Extensions.AI is a set of core .NET libraries (currently in preview) that provides a unified layer of C# abstractions for interacting with AI services. It is designed to work with the .NET ecosystem, including Semantic Kernel, to provide a consistent way to interact with Small Language Models (SLMs), Large Language Models (LLMs), and embeddings.
Key benefits include:
gpt-35-turbo model. The application reads a local file named benefits.md and sends a request to an Azure OpenAI Service to summarize the text into a few words.The Customer Support sample is a C# console application designed to demonstrate AI-driven workflows for customer service. It performs three primary tasks:
ManualIngestor.cs: Handles PDF data extraction, text chunking, embedding generation, and JSON storage.ProductManualService.cs: A storage service utilizing an IVectorStore to manage and query product manual embeddings.TicketSummarizer: An AI service that interfaces with models to summarize support tickets.These samples demonstrate how to integrate OpenAI models (gpt-3.5-turbo and dall-e-3) into .NET 8.0 console applications using the Semantic Kernel (SK) library. Semantic Kernel is an extensible SDK that allows you to build generative AI solutions and can be used with various model providers including OpenAI, Azure OpenAI, and Hugging Face.
Available sample scenarios include:
Microsoft.Extensions.AI.Abstractions and Microsoft.Extensions.AI NuGet packages. It is intended to help developers understand how to implement these abstractions or use them within their own applications.The Microsoft.Extensions.AI.Evaluation libraries provide tooling to evaluate the quality and efficacy of LLM responses. The ecosystem is divided into the following functional areas:
Microsoft.Extensions.AI.Evaluation defines the base types and interfaces.Microsoft.Extensions.AI.Evaluation.Quality provides evaluators for metrics like Relevance, Truth, Completeness, Fluency, Coherence, Retrieval, Equivalence, and Groundedness.Microsoft.Extensions.AI.Evaluation.Safety uses Azure AI Foundry Evaluation service to check for Protected Material, Groundedness Pro, Ungrounded Attributes, Hate and Unfairness, Self Harm, Violence, Sexual content, Code Vulnerability, and Indirect Attack.Microsoft.Extensions.AI.Evaluation.Reporting handles caching LLM responses and storing results. Microsoft.Extensions.AI.Evaluation.Reporting.Azure provides an implementation for storing these in Azure Storage containers.Microsoft.Extensions.AI.Evaluation.Console is a dotnet tool for managing evaluation data and generating reports.