调用LLM+RAG（外部文件+知识库）生成回答

llm-back/ruoyi-agent/src/main/java/com/ruoyi/agent/service/impl/McpServiceImpl.java

@@ -21,8 +21,8 @@ import java.util.Collections;
 @McpServerEndpoint(sseEndpoint = "/llm/mcp/sse")
 public class McpServiceImpl implements IMcpService {
 
-    @ToolMapping(description = "查看招标文档")
-    public String openDocument(@Param(description = "文件路径") String document) throws IOException {
+    @ToolMapping(description = "梳理招标文件服务要求")
+    public String getBidRequest(@Param(description = "文件路径") String document) throws IOException {
         try (XWPFDocument doc = new XWPFDocument(new FileInputStream(document))) {
             StringBuilder content = new StringBuilder();
             for (XWPFParagraph paragraph : doc.getParagraphs()) {
@@ -32,15 +32,7 @@ public class McpServiceImpl implements IMcpService {
         }
     }
 
-    @ToolMapping(description = "开场白")
-    public String opening(@Param(description = "智能体名称") String agentName) {
-        String content = "";
-        if (agentName.contains("技术"))
-            content = "我是投标文件写作助手，我将助您完成技术文件部分撰写。请上传招标询价文件，分析后将依招标服务要求，运用技术方案知识库，为您提供参考。";
-        return content;
-    }
-
-    @ToolMapping(description = "补全技术方案")
+    @ToolMapping(description = "撰写技术方案")
     public String writeTechnicalPlan(@Param(description = "文件路径") String document, @Param(description = "技术方案内容") String resultContent) throws IOException {
         // 创建临时输出文件路径
 

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/controller/McpController.java

@@ -2,15 +2,22 @@ package com.ruoyi.web.llm.controller;
 
 import com.alibaba.fastjson2.JSONObject;
 import com.ruoyi.common.core.controller.BaseController;
+import com.ruoyi.web.llm.service.ILangChainMilvusService;
+import dev.langchain4j.model.embedding.EmbeddingModel;
+import dev.langchain4j.model.embedding.onnx.bgesmallzhv15.BgeSmallZhV15EmbeddingModel;
+import io.milvus.client.MilvusServiceClient;
+import io.milvus.param.ConnectParam;
 import org.noear.solon.ai.chat.ChatModel;
 import org.noear.solon.ai.chat.ChatResponse;
 import org.noear.solon.ai.chat.message.AssistantMessage;
 import org.noear.solon.ai.chat.message.ChatMessage;
 import org.noear.solon.ai.mcp.client.McpClientProvider;
+import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.web.bind.annotation.GetMapping;
 import org.springframework.web.bind.annotation.RequestMapping;
 import org.springframework.web.bind.annotation.RestController;
 import org.springframework.web.multipart.MultipartFile;
+import reactor.core.publisher.Flux;
 
 import java.io.IOException;
 import java.util.*;
@@ -26,6 +33,17 @@ import java.util.*;
 @RequestMapping("/llm/mcp")
 public class McpController extends BaseController
 {
+    @Autowired
+    private ILangChainMilvusService langChainMilvusService;
+
+    private static final EmbeddingModel embeddingModel = new BgeSmallZhV15EmbeddingModel();
+
+    private static final MilvusServiceClient milvusClient = new MilvusServiceClient(
+            ConnectParam.newBuilder()
+                    .withHost("192.168.28.188")
+                    .withPort(19530)
+                    .build());
+
     /**
      * 同步问答
      * @return
@@ -54,6 +72,16 @@ public class McpController extends BaseController
         return assistantMessage;
     }
 
+    /**
+     * 调用LLM+RAG（外部文件）生成回答
+     */
+    @GetMapping("/answerWithDocument")
+    public Flux<AssistantMessage> answerWithDocumentAndCollection(String collectionName, String topicId, String chatId, String question) throws IOException
+    {
+        List<JSONObject> requests = langChainMilvusService.retrieveFromMilvus(milvusClient, embeddingModel, collectionName, question, 10);
+        return langChainMilvusService.generateAnswerWithDocumentAndCollection(embeddingModel, topicId, chatId, question, requests, "http://192.168.28.188:8000/v1/chat/completions");
+    }
+
 
     /**
      * 根据招标文件要求编写投标文件技术方案

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/service/ILangChainMilvusService.java

@@ -49,4 +49,10 @@ public interface ILangChainMilvusService {
      */
     public Flux<AssistantMessage> generateAnswerWithDocument(EmbeddingModel embeddingModel, String topicId, String chatId, String question, String llmServiceUrl) throws IOException;
 
+    /**
+     * 调用LLM+RAG（外部文件+知识库）生成回答
+     * @return
+     */
+    public Flux<AssistantMessage> generateAnswerWithDocumentAndCollection(EmbeddingModel embeddingModel, String topicId, String chatId, String question,  List<JSONObject> requests, String llmServiceUrl) throws IOException;
+
 }

llm-back/ruoyi-llm/src/main/java/com/ruoyi/web/llm/service/impl/LangChainMilvusServiceImpl.java

@@ -214,6 +214,40 @@ public class LangChainMilvusServiceImpl implements ILangChainMilvusService
         return generateAnswer(topicId, sb.toString(), llmServiceUrl);
     }
 
+    /**
+     * 调用LLM生成回答
+     */
+    @Override
+    public Flux<AssistantMessage> generateAnswerWithDocumentAndCollection(EmbeddingModel embeddingModel, String topicId, String chatId, String question, List<JSONObject> requests, String llmServiceUrl) throws IOException {
+        CmcDocument cmcDocument = new CmcDocument();
+        cmcDocument.setChatId(chatId);
+        List<CmcDocument> documentList = cmcDocumentService.selectCmcDocumentList(cmcDocument);
+        StringBuilder sb = new StringBuilder("问题: " + question + "\n\n").append("根据以下要求：\n\n");
+        for (JSONObject request : requests) {
+            sb.append("要求").append(": ")
+                    .append(request.getString("content")).append("\n\n");
+        }
+        sb.append("参考以下文件上下文回答问题：\n\n");
+        for (CmcDocument document : documentList) {
+            File profilePath = new File(RuoYiConfig.getProfile() + "/upload/rag/document/" + document.getPath());
+            List<TextSegment> segments = splitDocument(document.getPath(), profilePath);
+            List<Embedding> embeddings = embeddingModel.embedAll(segments).content();
+            InMemoryEmbeddingStore<TextSegment> embeddingStore = new InMemoryEmbeddingStore<>();
+            embeddingStore.addAll(embeddings, segments);
+            Embedding queryEmbedding = embeddingModel.embed(question).content();
+            EmbeddingSearchRequest embeddingSearchRequest = EmbeddingSearchRequest.builder()
+                    .queryEmbedding(queryEmbedding)
+                    .maxResults(1)
+                    .build();
+            String contexts = embeddingStore.search(embeddingSearchRequest).matches().get(0).embedded().text();
+            sb.append("文件").append(": ")
+                    .append(document.getPath()).append("\n\n")
+                    .append("上下文").append(": ")
+                    .append(contexts).append("\n\n");
+        }
+        return generateAnswer(topicId, sb.toString(), llmServiceUrl);
+    }
+
     /**
      * 检索知识库
      */