【二】springAi实现提问敏感词的过滤(自定义拦截器)

1. 一般情况，我们会涉及到一个重要的业务逻辑，就是对提问的敏感词进行过滤，在springAi中实现的方法也很简单,只需要在bean中添加一个SafeGuardAdvisor就可以了

@Bean(name = "openAiChatClient")
    public ChatClient openrouterChatClient(OpenAiChatModel openAiChatModel) {
        OpenAiChatOptions openAiChatOptions = new OpenAiChatOptions();
        openAiChatOptions.setModel("openai/gpt-5-mini");
        List<String> list = List.of("郭雲铨", "trump", "drug");
        SafeGuardAdvisor safeGuardAdvisor = new SafeGuardAdvisor(list,"你想害死我啊，别瞎打听",0);
        WriterAdvisor writerAdvisor = new WriterAdvisor();
        return ChatClient
                .builder(openAiChatModel).
                defaultAdvisors(safeGuardAdvisor).
                defaultOptions(openAiChatOptions)
                .build();
    }

这个例子就是当触发"郭雲铨", "trump", "drug"这些关键词时候，不调用大模型，直接返回违禁结果

• 这是正常提问，正常回答

• 然后提问违禁词

2. 这个里面关键实现了SafeGuardAdvisor，查看了一下文档，springAi是可以写拦截切面的（调用之前和调用之后）,写一个对用户提问重写的例子

package com.gyq.config;

import org.jetbrains.annotations.NotNull;
import org.springframework.ai.chat.client.ChatClientRequest;
import org.springframework.ai.chat.client.ChatClientResponse;
import org.springframework.ai.chat.client.advisor.api.AdvisorChain;
import org.springframework.ai.chat.client.advisor.api.BaseAdvisor;

import org.springframework.ai.chat.prompt.Prompt;
import org.springframework.ai.chat.prompt.PromptTemplate;

import java.util.Map;

public class WriterAdvisor implements BaseAdvisor {

    private static final String mutatePrompt = """
            你作为一个心血管方面的专家，请对我的问题进行深度理解,在你的回答中要有你职业的介绍
            {question}
            """;

    /**
     * Logic to be executed before the rest of the advisor chain is called.
     */
    @NotNull
    @Override
    public ChatClientRequest before(ChatClientRequest chatClientRequest, @NotNull AdvisorChain advisorChain) {
        String contents = chatClientRequest.prompt().getContents();
        String question = PromptTemplate.builder().template(mutatePrompt).build()
                .render(Map.of("question", contents));
        return chatClientRequest
                .mutate()
                .prompt(Prompt.builder().content(question).build())
                .build();
    }

    /**
     * Logic to be executed after the rest of the advisor chain is called.
     */
    @NotNull
    @Override
    public ChatClientResponse after(@NotNull ChatClientResponse chatClientResponse, @NotNull AdvisorChain advisorChain) {
        return chatClientResponse;
    }

    @Override
    public int getOrder() {
        return 0;
    }
}

然后接口调用新增WriterAdvisor

@GetMapping(value ="/openAi", produces = "text/stream;charset=UTF-8")
    public Flux<String> openAiChat(@RequestParam("userPrompt") String userPrompt) {
        return openAiChatClient
                .prompt().advisors(new WriterAdvisor())
                .user(userPrompt)
                .stream()
                .content();
    }

最终实现效果就是