[GitHub Global] Translate Vibe Coding 零基础教程/40 编程学习/04 AI 编程技术.md to en
This commit is contained in:
yupi-translate-app[bot]
committed by
GitHub
GitHub
parent
ce4e4dba1a
commit
a6a4995081
@@ -0,0 +1,281 @@
|
||||
# AI Programming Technology Beginner's Guide
|
||||
|
||||
> Master AI development frameworks and become a hot commodity in the job market
|
||||
|
||||
Hello, I'm programmer Yupi.
|
||||
|
||||
As programmers, we not only need to know how to use AI tools and leverage AI to develop projects, but also need to be able to independently develop AI projects and integrate AI capabilities into our own projects.
|
||||
|
||||
As the saying goes: **In the AI era, all traditional businesses are worth reshaping with AI.**
|
||||
|
||||
Therefore, many companies are now hiring programmers who can develop AI projects, which is also our opportunity. So what knowledge and technologies do we need to learn to become a hot commodity in the job market?
|
||||
|
||||
⭐️ Recommended video version: https://www.bilibili.com/video/BV1i9Z8YhEja
|
||||
|
||||
## 1. AI Development Frameworks
|
||||
|
||||
First, from a technical perspective, we need to learn mainstream AI development frameworks. Currently, the most popular ones in the Java direction are **Spring AI** and **LangChain4j**.
|
||||
|
||||
### Spring AI
|
||||
|
||||
[Spring AI](https://docs.spring.io/spring-ai/reference/getting-started.html) is an AI development framework officially launched by Spring. After two years of development, it officially released version 1.0 in May 2025.
|
||||
|
||||
|
||||
|
||||
The advantage of Spring AI lies in **its easier integration with the mainstream Java development framework Spring**, making it easier to get started. It provides many ready-made methods to help us improve the efficiency of developing AI applications, such as quickly connecting to large models, saving session context, connecting to vector databases for RAG, and more.
|
||||
|
||||
|
||||
|
||||
The core features of Spring AI include:
|
||||
|
||||
- Large model invocation capability: Unified interface supports multiple mainstream large models (OpenAI, Claude, Tongyi Qianwen, etc.)
|
||||
- Prompt engineering: Provides Prompt and PromptTemplate classes for easy management of prompts
|
||||
- Session memory: One line of code to enable conversation memory, automatically handling context
|
||||
- RAG retrieval-augmented generation: Full RAG process support, including document loading, vector storage, retrieval optimization
|
||||
- Tool invocation: Quickly define tools through annotations, allowing AI to call external services
|
||||
- MCP support: Easily access and develop MCP services
|
||||
|
||||
For example, using Spring AI to call a large model requires only a few lines of code:
|
||||
|
||||
```java
|
||||
// Using Spring AI to call a large model
|
||||
@Bean
|
||||
public ChatClient chatClient(ChatModel chatModel) {
|
||||
return ChatClient.builder(chatModel).build();
|
||||
}
|
||||
|
||||
public String doChat(String message) {
|
||||
return chatClient.prompt(message).call().content();
|
||||
}
|
||||
```
|
||||
|
||||
Without Spring AI, you would need to write HTTP requests and parse responses yourself, which is much more cumbersome.
|
||||
|
||||
### Spring AI Alibaba
|
||||
|
||||
[Spring AI Alibaba](https://java2ai.com/) is a domestic version launched by Alibaba based on Spring AI, specifically optimized for the domestic AI ecosystem.
|
||||
|
||||
Its advantages include:
|
||||
|
||||
- Better support for domestic large models (Tongyi Qianwen, Baidu Wenxin Yiyan, etc.)
|
||||
- Provides Chinese documentation and technical support
|
||||
- Optimized for domestic network environments
|
||||
- Supported by Alibaba Cloud's ecosystem
|
||||
|
||||
If you mainly use domestic AI services, Spring AI Alibaba would be a better choice.
|
||||
|
||||
### LangChain4j
|
||||
|
||||
[LangChain4j](https://docs.langchain4j.dev/intro) is another mainstream Java AI development framework, characterized by **greater flexibility, more suitable for developing complex agents**.
|
||||
|
||||
For example, when developing an intelligent document analysis system, using LangChain4j, the agent can automatically read document content, call a search engine to obtain relevant background knowledge, and then combine document information with external knowledge based on task requirements to generate an analysis report.
|
||||
|
||||
The core features of LangChain4j include:
|
||||
|
||||
- AI Service: Declarative development, quickly building AI services through annotations
|
||||
- Session memory: Supports multiple session memory strategies and persistence
|
||||
- Structured output: Automatically converts AI output into Java objects
|
||||
- RAG support: Full RAG process, supports multiple vector databases
|
||||
- Tool invocation: Flexible tool definition and invocation mechanism
|
||||
- Guardrails: Input/output interceptors, enhancing security
|
||||
|
||||
For example, using LangChain4j's AI Service:
|
||||
|
||||
```java
|
||||
@AiService
|
||||
public interface AiCodeHelperService {
|
||||
@SystemMessage(fromResource = "system-prompt.txt")
|
||||
String chat(String userMessage);
|
||||
}
|
||||
```
|
||||
|
||||
Just define the interface and annotations, and the framework will automatically generate the implementation class, which is very convenient.
|
||||
|
||||
### How to Choose a Framework?
|
||||
|
||||
| Scenario | Recommended Framework | Advantages |
|
||||
| ----------------- | ---------------------------- | -------------------------- |
|
||||
| Java enterprise applications | Spring AI | Seamless integration with Spring ecosystem |
|
||||
| Domestic AI services | Spring AI Alibaba | Better support for domestic large models |
|
||||
| Agent development | LangChain4j | Complete Agent toolchain |
|
||||
| Complex workflows | LangGraph (advanced) | Visual orchestration |
|
||||
|
||||
My suggestion is to **learn both, starting with Spring AI, then learning LangChain4j will be easier**. Many concepts and usages are similar, and mastering one will make it easier to pick up the other.
|
||||
|
||||
## 2. AI Integration
|
||||
|
||||
The prerequisite for developing AI applications is having a large model, but large models require computing power to run, and computing power costs money. Where can we get large models?
|
||||
|
||||
There are two methods: using AI cloud services or deploying large models locally.
|
||||
|
||||
### AI Cloud Services
|
||||
|
||||
AI cloud services are AI large models deployed by other enterprises, provided to us through API interfaces, charged based on usage.
|
||||
|
||||
For example, Alibaba Cloud Bailian, Volcano Engine, Silicon Flow, OpenAI, etc.
|
||||
|
||||
|
||||
|
||||
What we programmers need to focus on is:
|
||||
|
||||
1. How to access cloud services through APIs?
|
||||
2. How to use AI cloud services to create agents and configure parameters?
|
||||
3. How to choose the right cloud service? This requires attention to the pricing models and service quality of various cloud services
|
||||
4. How to use cloud services at lower costs and more stably? This requires us to learn Prompt engineering and high-availability technologies
|
||||
|
||||
### Local Deployment of Large Models
|
||||
|
||||
Local deployment of large models is also a necessity for many enterprises, as data does not need to be uploaded to the cloud, effectively ensuring data security and privacy, especially suitable for industries extremely sensitive to data security such as healthcare and finance.
|
||||
|
||||
Local deployment of large models is actually not difficult, just use the [Ollama tool](https://ollama.com/) to deploy various mainstream open-source models with one click.
|
||||
|
||||
|
||||
|
||||
However, the difficulty of deploying large models is not technical, but mainly the lack of computing power! Otherwise, I would have deployed a Yupi large model for our team's [Programming Navigation](https://codefather.cn) and [Interview Duck](https://www.mianshiya.com).
|
||||
|
||||
## 3. AI Domain Business
|
||||
|
||||
AI business development in enterprises is not just about AI conversations; we also need to master several more complex business developments, such as RAG knowledge bases, multimodal, MCP services, and ReAct agents.
|
||||
|
||||
### RAG Knowledge Base
|
||||
|
||||
Many companies have their own business knowledge and documents, and will build their own Q&A systems or customer service, which requires the use of RAG retrieval-augmented generation technology.
|
||||
|
||||
First, use a text embedding model to convert various enterprise documents into vectors and store them in a vector database; when users ask questions, the system retrieves relevant vector data in the vector database, finds the most similar document fragments, and inputs them into the large model along with the question. This way, the large model can answer based on real enterprise data, more accurately fitting the actual situation.
|
||||
|
||||
|
||||
|
||||
There is a lot to learn about RAG, such as mainstream vector databases Milvus and PGVector, document extraction/conversion/loading, index construction, query strategy optimization, etc. **This is also a key focus in AI enterprise interviews!**
|
||||
|
||||
### Multimodal
|
||||
|
||||
Multimodal is also a mainstream AI business scenario, integrating multiple different types of data modalities such as text, images, audio, and video, thereby improving product usability and creating more creative features.
|
||||
|
||||
For example, building an intelligent shopping guide system, where customers can input text descriptions of desired products, the system can also recognize product images uploaded by customers, and even understand shopping needs expressed through voice.
|
||||
|
||||
|
||||
|
||||
To develop multimodal applications, we need to learn modal conversion technologies, such as text-to-speech (TTS), speech-to-text (STT), optical character recognition (OCR), etc. However, these have ready-made tool libraries or cloud services, just master the calling methods.
|
||||
|
||||
### MCP Services
|
||||
|
||||
MCP (Model Context Protocol) can be understood as various services provided to AI, enabling AI to achieve more powerful functions.
|
||||
|
||||
|
||||
|
||||
How to integrate others' MCP services into your project to enhance your project's capabilities; and how to develop your own MCP services for others' projects to use, are essential to learn.
|
||||
|
||||
Now, using development frameworks like Spring AI, you can develop MCP services, and even experts have created a [website](http://mcpify.ai/) that can create your own MCP service with one sentence, which is really cool.
|
||||
|
||||
|
||||
|
||||
### ReAct Agent
|
||||
|
||||
ReAct is a development paradigm for building agents, aiming to create agents that can autonomously take actions based on reasoning results.
|
||||
|
||||
Its development process involves task planning, tool invocation, interactive I/O, exception handling, etc. Especially tool invocation, which can be achieved through Function Call or MCP to implement functions like weather query, file reading/writing, webpage running, information retrieval, terminal command execution, etc.
|
||||
|
||||
|
||||
|
||||
Take developing a video website as an example, when the user gives the instruction "Help me develop a Dilidili video website and deploy it online", the agent will first deeply understand the task content, reason through a series of execution steps, including clarifying requirements, designing solutions, building frameworks, generating code, deploying online, etc. Next, the agent will call the corresponding tools to execute these actions.
|
||||
|
||||
|
||||
|
||||
## 4. AI Toolchain
|
||||
|
||||
Finally, there are some platforms, tools, and libraries that we might use when developing AI projects.
|
||||
|
||||
### Low-Code Platforms
|
||||
|
||||
For example, the mainstream low-code AI development platform [Dify](https://dify.ai/), which allows us to build our own AI agents through drag-and-drop, create knowledge bases and import our own documents, build complex workflows, etc. Even if you don't know how to code, you can create complex AI applications with it.
|
||||
|
||||
|
||||
|
||||
### Tool Libraries
|
||||
|
||||
There are also some tool libraries that are useful when developing AI agents, such as:
|
||||
|
||||
- Apache Tika: A powerful file parser tool library, supports parsing various documents like PDF, Word, Excel, PowerPoint, etc.
|
||||
- Playwright: A tool library for simulating browser behavior, useful when AI needs to run web pages, scrape web data, automate testing, etc.
|
||||
- JSON format parsing libraries: GSON and Kryo
|
||||
- HTML document parsing library: jsoup
|
||||
|
||||
These libraries have almost no learning cost, just refer to the documentation when needed.
|
||||
|
||||
### Deployment Tools
|
||||
|
||||
Projects ultimately need to be deployed online, so we also need to master efficient deployment tools. If you are learning individually and want to quickly deploy your own small AI applications, you can try the following platforms:
|
||||
|
||||
- [Vercel](https://vercel.com/): Suitable for front-end application deployment platforms, supports automatic building, online browsing, CDN distribution, and even provides free accessible domains
|
||||
- [Sealos](https://sealos.io/): Cloud-native application management platform, supports Kubernetes cluster management
|
||||
- [Railway](https://railway.com/): Allows developers to easily deploy Docker containers without worrying about server configuration and maintenance
|
||||
|
||||
Of course, to quickly deploy services, Docker containerization technology is also essential, like an APP installation package, easily distributing and deploying your applications.
|
||||
|
||||
|
||||
|
||||
## 5. Recommended Learning Resources
|
||||
|
||||
How about it? There's quite a lot to learn. Don't worry, I'm also continuously learning these contents and will continue to share them with you.
|
||||
|
||||
### 1. AI Learning Path
|
||||
|
||||
The complete AI large model application development learning path can be obtained at [Programming Navigation](https://www.codefather.cn/course/1789189862986850306/section/1912024009574629377):
|
||||
|
||||
URL: https://www.codefather.cn/learn
|
||||
|
||||
|
||||
|
||||
### 2. AI Project Practice
|
||||
|
||||
On [Programming Navigation](https://www.codefather.cn), I have led everyone through multiple sets of AI project tutorials, covering almost all the technologies mentioned above:
|
||||
|
||||
- AI Programming Assistant: LangChain4j framework introduction, practical dialogue memory, structured output, RAG, tool invocation, MCP, SSE, etc.
|
||||
- AI Super Agent: Spring AI framework introduction, practical AI love master application + autonomous planning super agent
|
||||
- AI Zero-Code Application Generation Platform: LangChain4j AI agent, LangGraph4j workflow, microservice architecture
|
||||
- AI Q&A Application Platform: React cross-platform mini-program, Vue3 AI application, database sharding, SSE real-time push
|
||||
|
||||
These projects are all enterprise-level real projects, and you can directly write them into your resume after completion.
|
||||
|
||||
### 3. Open Source Projects
|
||||
|
||||
I have also open-sourced many AI application development projects, sharing them with everyone:
|
||||
|
||||
- AI Application Generation Platform: https://github.com/liyupi/yu-ai-code-mother
|
||||
- AI Super Agent: https://github.com/liyupi/yu-ai-agent
|
||||
- AI Literature Reading Assistant: https://github.com/liyupi/literature-assistant
|
||||
- AI Knowledge Base: https://github.com/liyupi/ai-guide
|
||||
|
||||
### 4. AI Knowledge Base
|
||||
|
||||
In my free open-source [AI Knowledge Base](https://github.com/liyupi/ai-guide), I have collected the latest and most comprehensive AI knowledge to help everyone better adapt to the arrival of the AI era.
|
||||
|
||||
URL: https://ai.codefather.cn
|
||||
|
||||
|
||||
|
||||
In addition to various tutorial materials, I also focus on sharing many specific application scenarios of AI tools, such as integrating office software to improve efficiency, helping you do self-media, AI batch video production, etc., hoping to help everyone draw inferences and find new ideas.
|
||||
|
||||
## Final Words
|
||||
|
||||
AI technology is evolving rapidly, and the requirements for programmers are also constantly increasing. **AI-related knowledge is no longer just for algorithm engineers to understand, but a basic skill that every programmer must master.**
|
||||
|
||||
Whether you are a front-end, back-end, or full-stack developer, you need to understand the basic concepts and practical methods of AI application development.
|
||||
|
||||
Because in the future of software development, AI will be everywhere.
|
||||
|
||||
If you ask me if AI will eliminate programmers?
|
||||
|
||||
My answer is still "yes". Because programmers themselves need continuous learning and practice to maintain competitiveness, as long as everyone can learn the knowledge I mentioned, pay more attention to the forefront of AI, I believe AI will not take away our jobs as programmers, but become our lever to transform the world.
|
||||
|
||||
## Recommended Resources
|
||||
|
||||
1) Yupi AI Navigation Website: [AI Resource Collection, Latest AI News, Free AI Tutorials](https://ai.codefather.cn)
|
||||
|
||||
2) Programming Navigation Learning Circle: [Learning Path, Programming Tutorials, Practical Projects, Job Hunting Guide, Q&A](https://www.codefather.cn)
|
||||
|
||||
3) Programmer Interview Eight-Part Essay: [Internship/Campus Recruitment/Social Recruitment High-Frequency Test Points, Enterprise Real Questions Analysis](https://www.mianshiya.com)
|
||||
|
||||
4) Programmer Resume Writing Tool: [Professional Templates, Rich Examples, Direct to Interview](https://www.laoyujianli.com)
|
||||
|
||||
5) 1-on-1 Mock Interview: [Internship/Campus Recruitment/Social Recruitment Interview Essential for Getting Offers](https://ai.mianshiya.com)
|
||||
Reference in New Issue
Block a user