Location>code7788 >text

AI and .NET Technology Practical Series (9): Summary Chapter ── Exploring the AI ​​Ecology of .NET Development: Tools, Library and Future Trends

Popularity：155 ℃/2025-03-24 10:25:55

1. Introduction

This article is the last article in this series, aiming to comprehensively explore the tools, libraries, frameworks and resources related to AI in the .NET ecosystem, and help developers understand how to develop AI applications in the .NET environment. We will analyze Microsoft's AI strategy, core tools and libraries (such as Azure AI SDK and Semantic Kernel), new features of .NET 9, and community resources and future trends, providing a clear roadmap for .NET developers overall.

2. Overview of .NET AI Ecology

The current situation of the .NET AI ecosystem

Microsoft has long been a leader in AI technology, with its AI strategy covering multiple levels from cloud services to developer tools. The Azure AI platform provides a wealth of services, including machine learning, cognitive services and pre-trained models, through which developers can quickly integrate AI capabilities into their applications. In addition, Microsoft has launched , an open source machine learning framework designed for .NET developers, allowing developers to build and deploy machine learning models in the familiar .NET ecosystem without switching to other languages ​​such as Python. This strategic layout embodies Microsoft's vision to popularize AI and integrate it into developers' daily work.

Components of the .NET AI ecosystem

.NET AIThe ecosystem is composed of multiple core components, each of which plays an important role in the development of AI applications:

  • : A cross-platform open source machine learning framework that enables developers to build custom machine learning models using C# or F#.
  • Azure AI SDK: Provides access to Azure AI services, including Azure OpenAI, Azure Cognitive Services, etc., to help developers easily integrate cloud AI capabilities.
  • Semantic Kernel: An open source SDK focusing on building applications based on large language models (LLM) and closely integrates with the .NET ecosystem.
  • : A preview package that provides a unified AI service interface to simplify the integration of AI functions.

Together, these components form a powerful ecosystem that enables .NET developers to flexibly deal with AI tasks ranging from simple prediction to complex deep learning.

3. Core tools and libraries

It is an open source machine learning framework launched by Microsoft, designed for .NET developers. It lowers the barriers to entry for machine learning, allowing developers to build and deploy models without a deep math or AI background. Supports a variety of tasks, including classification, regression, clustering, recommendation system and anomaly detection.

Functions and features

  • Ease of use: Through high-level APIs, developers can quickly define and train models.
  • Integrate with .NET: Completely based on C# or F#, no need to learn new languages.
  • Flexible data processing: Supports multiple data sources such as CSV, JSON and databases.
  • Model explanatory: Built-in tools help developers understand the prediction logic of models.

Use

Here is a typicalDevelopment process:

  1. Data preparation: Load the dataset and preprocess it, such as normalization or missing values ​​filling.
  2. Model training: Select algorithms (such as decision trees or neural networks) to train models.
  3. Model evaluation: Use test data to evaluate the accuracy or error of the model.
  4. deploy: Save and integrate models into .NET applications, such as Web APIs or desktop programs.

For example, developers can use to build a model that predicts housing prices, which can complete data loading, model training, and prediction in just a few lines of C# code.

Azure AI SDK

Azure AI SDKis a set of tools and libraries provided by Microsoft to access Azure's cloud AI services. These services include Azure OpenAI, Azure Cognitive Services, and Azure Machine Learning, providing developers with powerful pre-built AI capabilities.

Functions and advantages

  • Diversified AI services: Covering natural language processing (NLP), computer vision, speech recognition and other fields.
  • Easy to integrate: Provides REST API and SDK to support cross-language development.
  • Enterprise-level support: High availability, security and compliance.

Supported AI services

  • Azure OpenAI: Provides access to OpenAI models such as GPT-4 to generate text or answer questions.
  • Azure Cognitive Services: Including functions such as text analysis, image recognition, voice to text.
  • Azure Machine Learning: Supports the training and deployment of custom models.

passAzure AI SDK, developers can add voice recognition or sentiment analysis to the app in minutes without building a model from scratch.

Semantic Kernel

Semantic KernelIt is an open source SDK designed to help developers build AI-based applications, especially those related to large language models. It is developed by Microsoft and integrates seamlessly with the .NET ecosystem.

Concepts and applications

  • AI Orchestration: Supports the combination of multiple AI services into complex workflows.
  • Plug-in system: Allow developers to create custom plugins and extend functions.
  • Language Support: Written in C#, suitable for .NET developers.

For example, developers can useSemantic KernelBuild an intelligent customer service system, pass user input to a model to generate replies, and handle specific needs in combination with custom logic.

ONNX Runtime: High-performance cross-platform model deployment

ONNX RuntimeIt is a high-performance inference engine developed by Microsoft for deploying deep learning models on multiple platforms. It supports models trained by frameworks such as TensorFlow, PyTorch, etc. through the ONNX format, providing cross-platform, high-performance inference capabilities, suitable for cloud and edge devices.

Architecture and working principle

ONNX RuntimeIt adopts a modular architecture, including model loading, graph optimization and execution engine, and supports a variety of execution providers such as CPU and GPU. After loading the ONNX model, it optimizes the computational graph and performs inference efficiently to adapt to different hardware environments.

Use in .NET

Developers can useThe NuGet package uses ONNX Runtime in .NET. The steps include loading the model, preparing the input data, running inference, and processing the output. For example:

var model = new InferenceSession("");
var inputTensor = new DenseTensor<float>(new[] { 1, 3, 224, 224 });
var inputs = new List<NamedOnnxValue> { ("input", inputTensor) };
var results = (inputs);
var output = ().AsTensor<float>();

Performance optimization

ONNX RuntimeIt supports graph optimization, hardware acceleration (such as GPU) and model quantization, significantly improves inference efficiency, and is suitable for resource-constrained scenarios.

Integrate with other tools

It can be integrated with, deploy training models, or combined with Azure AI services to enable cloud inference and enhance the flexibility of .NET AI applications.

ONNX RuntimeProvides a powerful tool for .NET developers to accelerate AI model deployment and application development.

4.

Provides a unified abstraction layer for .NET developers to integrate AI services into applications and libraries. It is built on the dependency injection and configuration mode of .NET, seamlessly connected with existing .NET applications.

Design goals

Its core goal is to simplify the calling process of AI services, so that developers can use their functions without having to understand the specific implementation details of each AI service.

  • Unified interface: Provide a consistent API for different AI services to reduce learning and usage costs.
  • flexibility: Support switching between different AI services without major code modification.
  • Ecological collaboration: Package authors in the .NET ecosystem are encouraged to integrate this abstraction into their respective libraries to promote consistency and interoperability.

In the .NET AI ecosystem, it is similar to a bridge that connects applications with diverse AI services, becoming a capable assistant for developers to build intelligent applications.

Main functions

  • AI service call: Supports common AI tasks such as text generation, image processing, embedding generation, etc.
  • Request Management: Processing request parameters and response data of AI services, including streaming responses.
  • Middleware support: Allow developers to customize the processing logic of AI requests, such as logging or caching.

Through these functions,Enable developers to access local or cloud AI services in a consistent manner.

Main interfaces and APIs

Mainly includingandThe core of these two libraries lies in their abstraction layer, which mainly includes the following interfaces and classes:

  • IChatClient: Used to interact with chat models for intelligent dialogue and text generation.
using ;

IChatClient client = new SampleChatClient(
    new Uri(""), "target-ai-model");

var response = await ("What is AI?");

();
  • IEmbeddingGenerator: Embed vectors used to generate text to support features such as semantic search and recommendation.
using ;

IEmbeddingGenerator<string, Embedding<float>> generator =
    new SampleEmbeddingGenerator(
        new Uri(""), "target-ai-model");

foreach (var embedding in await (["What is AI?", "What is .NET?"]))
{
    ((", ", ()));
}

How to simplify AI service integration

Simplify the integration of AI services by:

  • Hide implementation details: Different AI services (such as OpenAI and Azure AI) have their own APIs and parameters.Mask these differences through abstraction layers.
  • Support service switching: Developers can switch to other AI services by simply changing the configuration, without rewriting business logic.
  • Middleware flexibility: Allows adding custom logic to request processing, such as authentication or error retry.
  • Compatible with .NET ecosystem: Follow .NET's dependency injection and configuration mode to facilitate integration into existing projects.

Benefits of Unified Abstraction Layer

The unified abstraction layer brings the following benefits:

  • Lower the learning curve: Developers only need to master a set of APIs to use multiple AI services, without delving into the documentation and implementation details of each AI service.
  • Easy to maintain: When the AI ​​service is updated, you only need to update the underlying implementation and the business logic remains unchanged.
  • Code reusable: The same code logic can be applied to different AI services to reduce duplicate work.
  • Flexible switch: When replacing an AI service provider in a project, you only need to adjust the configuration without refactoring.
  • Supports mixed use: Multiple AI services can be combined in the same application, such as local models for low-latency tasks, and cloud models for complex computing.
  • Unified error handling: Centrally manage exceptions and errors through abstraction layers.
  • Test-friendly: Supports simulation and unit testing.
❝

A project using OpenAI only needs to change the service configuration without modifying the core code.

New AI-related features in .NET 9

AI building modules

.NET 9passandPackages introduce a unified C# abstraction layer, simplifying interaction with AI services. These services include small and large language models (SLMs and LLMs), embedding, vector storage, and middleware, providing developers with flexible AI integration tools.

TensorPrimitives and Tensor

  • TensorPrimitives: The method has been expanded from 40 to nearly 200, supporting similarMathandINumber<T>numerical operation, oriented to span data. Added SIMD optimization to improve performance, generic reload supports any type of implementation of specific interfacesT。
  • **Tensor**: Based onTensorPrimitives, provides efficient mathematical operations and supports copy-free interoperability with AI libraries such as TorchSharp, ONNX Runtime. Provides efficient data operations such as indexing and slicing, which are experimental features in .NET 9.

4.0 enhances the AI ​​development experience:

  • Added MLContext programming configuration options.
  • Supports loading of ONNX models in streams.
  • Improved DataFrame functionality.
  • Added tokenizer support.
  • (Experimental) TorchSharp transplants the Llama and Phi model families.
  • (Experimental) CausalLM Pipeline API.

Tokenizers

The library provides support for text encoding and decoding for context management, costing, and preprocessing in AI scenarios. The new version supports multiple models:

  • Tiktoken: Supports GPT (3, 3.5, 4, 4o, o1) and Llama3.
  • Llama: Based on SentencePiece, supports Llama and Mistral.
  • CodeGen: Support code generation models such as codegen-350M-mono.
  • Phi2: Based on CodeGen, supports Microsoft Phi2.
  • WordPiece and Bert: Supports Bert models such as optimum--all-MiniLM-L6-v2.

6. Comparison of .NET AI main tools and libraries

The following table lists the commonly used AI SDKs and libraries in .NET, comparing their characteristics, applicable scenarios, and advantages and disadvantages:

Tools/Library Features Applicable scenarios Pros and cons
Easy to use, tightly integrated with .NET Custom machine learning models Advantages: Simple and easy to use; Disadvantages: Limited functions
Azure AI SDK Rich cloud AI services Cloud AI feature integration Advantages: Powerful; Disadvantages: Relying on the Internet
Supports TensorFlow models Deep Learning Applications Advantages: High flexibility; Disadvantages: Steep learning curve
CNTK Powerful deep learning toolkit Complex deep learning tasks Advantages: Excellent performance; Disadvantages: Insufficient documentation
Focus on image and audio processing Image recognition, audio analysis Advantages: Strong professionalism; Disadvantages: Small community

These tools each have their own focus, and developers can choose the right tools according to project needs. For example, it is suitable for rapid prototyping, but is more suitable for advanced applications that require deep learning.

7. Development practices and cases

AI application development process

Developing AI applications in .NET usually involves the following steps:

  1. Requirements Analysis: Definite AI tasks such as prediction, classification, or generating content.
  2. Data preparation: Collect data and clean and pre-process.
  3. Model selection and training: Select the algorithm and train the model.
  4. Model evaluation: Test model performance and adjust parameters.
  5. Deployment and integration: Deploy the model to the production environment and integrate it into the application.

Application Cases

  • Natural Language Processing: Use Azure Cognitive Services to analyze the sentiment of customer reviews and help businesses optimize services.
  • Image recognition: Use training a model to identify defective products on the production line.
  • ……

8. Community and learning resources

Learning Resources

Official learning resources

  • Microsoft Learn: Provides AI tutorials from introductory to advanced, such asdotnet-ai-ecosystem。
  • ai-samples：/dotnet/ai-samples/tree/main/src/microsoft-extensions-ai-evaluation/api

GitHub

On GitHubdotnet-ai-resourcesA repository is a valuable collection of resources that contain libraries, tools, and examples related to .NET AI development.

Main content
  • Recommended library: Such as OllamaSharp, Azure AI Inference SDK, etc.
  • Sample Project: Provides code examples from simple classification to complex deep learning.
  • Learning Resources: Links to official documentation and community tutorials.

Developer Community

SciSharp Stack

  • address：/SciSharp/

The .NET AI community is highly active and technical seminars will also be held. Developers can actively participate in these activities by following community dynamics to obtain the latest updates and technical support.

9. Future trends and prospects

Development trends of AI technology

AI technology is developing in the direction of deep learning, generative AI and reinforcement learning. These trends require higher computing power and more flexible development tools.

.NET's Opportunities in the AI ​​field

With the improvement of the .NET ecosystem, developers can leverage their cross-platform capabilities and rich toolsets to gain a foothold in the AI ​​field. Mastering .NET AI skills will bring more opportunities to career development.

10. Summary

As a powerful development platform, .NET provides rich tools and resources in the field of AI. From on-premises machine learning to the cloud capabilities of Azure AI, to innovative applications of Semantic Kernel, to unified integration, .NET developers have all the necessary tools to build smart applications.

With the release of .NET 9, its capabilities in AI development will be further improved.

Developers should explore the .NET AI ecosystem in depth, actively participate in community activities, and start building their own AI projects.

11. References

  • Microsoft Learn: .NET AI Ecosystem
  • ai-samples
  • Microsoft Learn: AI Extensions
  • GitHub: dotnet-ai-resources
  • What's new in .NET 9
  • Groove Technology: AI for .NET Developers
  • Restack: AI Libraries for C#
<Previous Next>