RAG（Retrieval-Augmented Generation）

// 使用 Semantic Kernel 實作 RAG
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Connectors.OpenAI;
using Microsoft.SemanticKernel.Memory;
 
// 建立索引
var memoryBuilder = new MemoryBuilder()
    .WithOpenAITextEmbeddingGeneration("text-embedding-3-small", apiKey)
    .WithQdrantMemoryStore("http://localhost:6333", 1536)
    .Build();
 
// 存入文件（自動切塊和 embedding）
await memoryBuilder.SaveInformationAsync("docs",
    id: "doc-001",
    text: "微服務架構是將應用拆分為多個獨立部署的服務...",
    additionalMetadata: "source:architecture-guide.md");
 
// 查詢
var results = memoryBuilder.SearchAsync("docs",
    query: "什麼是微服務？",
    limit: 3,
    minRelevanceScore: 0.7);
 
// RAG — 組合 context 送入 LLM
var context = string.Join("\n\n", await results
    .Select(r => r.Metadata.Text).ToListAsync());
 
var prompt = $@"根據以下參考資料回答問題。如果資料中沒有答案，請說「我不確定」。
 
參考資料：
{context}
 
問題：什麼是微服務？";
 
var answer = await kernel.InvokePromptAsync(prompt);

import { OpenAI } from "openai";
import { QdrantClient } from "@qdrant/js-client-rest";
 
const openai = new OpenAI();
const qdrant = new QdrantClient({ url: "http://localhost:6333" });
 
// Step 1: Chunking
function chunkText(text: string, maxTokens = 300, overlap = 50): string[] {
  const words = text.split(/\s+/);
  const chunks: string[] = [];
  for (let i = 0; i < words.length; i += maxTokens - overlap) {
    chunks.push(words.slice(i, i + maxTokens).join(" "));
  }
  return chunks;
}
 
// Step 2: 產生 Embedding 並存入向量 DB
async function indexDocument(docId: string, text: string) {
  const chunks = chunkText(text);
  for (let i = 0; i < chunks.length; i++) {
    const embedding = await openai.embeddings.create({
      model: "text-embedding-3-small",
      input: chunks[i],
    });
    await qdrant.upsert("documents", {
      points: [{
        id: `${docId}-${i}`,
        vector: embedding.data[0].embedding,
        payload: { text: chunks[i], docId, chunkIndex: i },
      }],
    });
  }
}
 
// Step 3: 查詢 + 生成
async function ragQuery(question: string): Promise<string> {
  // Retrieve
  const queryEmbedding = await openai.embeddings.create({
    model: "text-embedding-3-small",
    input: question,
  });
  const results = await qdrant.search("documents", {
    vector: queryEmbedding.data[0].embedding,
    limit: 5,
  });
  const context = results.map(r => r.payload?.text).join("\n\n");
 
  // Generate
  const response = await openai.chat.completions.create({
    model: "gpt-4o",
    messages: [
      { role: "system", content: "根據提供的參考資料回答問題。引用資料來源。如果資料不足，明確表示。" },
      { role: "user", content: `參考資料：\n${context}\n\n問題：${question}` },
    ],
  });
  return response.choices[0].message.content ?? "";
}

from langchain_community.vectorstores import Qdrant
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
 
# Step 1: Chunking
splitter = RecursiveCharacterTextSplitter(
    chunk_size=500,
    chunk_overlap=50,
    separators=["\n\n", "\n", "。", ".", " "]
)
chunks = splitter.split_text(document_text)
 
# Step 2: 存入向量 DB
embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
vectorstore = Qdrant.from_texts(
    texts=chunks,
    embedding=embeddings,
    url="http://localhost:6333",
    collection_name="documents"
)
 
# Step 3: RAG Chain
retriever = vectorstore.as_retriever(search_kwargs={"k": 5})
llm = ChatOpenAI(model="gpt-4o")
 
prompt = ChatPromptTemplate.from_template("""
根據以下參考資料回答問題。如果資料中沒有答案，請說「我不確定」。
引用你參考的段落。
 
參考資料：
{context}
 
問題：{question}
""")
 
rag_chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | llm
)
 
answer = rag_chain.invoke("什麼是微服務？")
print(answer.content)