fastAPI 심화

- Chart.js - pdf, csv 파일 업로드 후 데이터 정제하여 llm으로 처리 후 결과 도출 - sqlite로 데이터 저장 - ORM - SQLAlchemy
2026-06-16 18:03:02 +09:00
parent ccfdac1286
commit 06eb3c57ab
43 changed files with 1912 additions and 39 deletions
@@ -1,25 +1,33 @@
 from fastapi import APIRouter, Request, UploadFile
 from backend.services.llm_service import question_and_answer
 from backend.schemas.basic_schema import QuestionRequest
-from backend.services.rag_service import upload_document
+from backend.services.rag_service import upload_document, rag_chat, rag_chat_stream
+from fastapi.responses import StreamingResponse
 router = APIRouter(prefix="/api")

 # http://127.0.0.1:8000/api/question
@router.post("/question")
-async def question(req:QuestionRequest):
+async def question(req: QuestionRequest):
    answer = question_and_answer(req.question)

    return {"message" : answer}

 # http://127.0.0.1:8000/api/rag/upload
@router.post("/rag/upload")
-async def fileUpload(file:UploadFile):
+async def file_Upload(file: UploadFile):
    # 서비스 호출
    return upload_document(file)


 # http://127.0.0.1:8000/api/rag/question
-
@router.post("/rag/question")
-async def question():
-    pass
+async def question(req: QuestionRequest):
+    answer = rag_chat(req.question)
+
+    return {"message" : answer}
+
+@router.post("/rag/question/stream")
+async def question_stream(req: QuestionRequest):
+    answer = rag_chat_stream(req.question)
+
+    return StreamingResponse(rag_chat_stream(req.question), media_type="text/plain")
@@ -1,6 +1,90 @@
-# pdf 업로드 => 분할 => 인덱스 생성
-def upload_document(file):
-    pass
+import os

+from langchain_community.document_loaders import PyPDFLoader
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+from langchain_community.vectorstores import FAISS
+from langchain_core.runnables import RunnablePassthrough
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.prompts import ChatPromptTemplate
+
+from backend.ai.embedding import watson_embedding
+from backend.ai.llm import watson_llm
+
+UPLOAD_PATH = "uploads"
+
+def upload_document(file):
+    # file 저장
+    file_path = os.path.join(UPLOAD_PATH, file.filename)
+
+    with open(file_path, "wb") as f:
+        f.write(file.file.read())
+
+    # pdf 업로드 => 분할 => 인덱스 생성
+    # pdf 로드
+    loader = PyPDFLoader(file_path)
+    docs = loader.load()
+
+    splitter = RecursiveCharacterTextSplitter(chunk_size=300, chunk_overlap=30)
+    chunks = splitter.split_documents(docs)
+
+    faiss_store = FAISS.from_documents(documents=chunks, embedding=watson_embedding)
+    faiss_store.save_local("./db/vectorstore")
+
+    return {"message": "업로드 성공"}

 # 질문 => 유사도 검색 => 문서 => llm 답변 생성
+def rag_chat(question: str):
+
+    faiss_store = FAISS.load_local("./db/vectorstore", watson_embedding, allow_dangerous_deserialization=True)
+
+    retriever = faiss_store.as_retriever(search_kwargs={"k": 3})
+
+    ### LLM
+    # 1. prompt
+    message = """\
+    당신은 PDF 기반 RAG AI 입니다.
+    다음 문서를 참고해서 질문에 답변하세요.
+
+    문서:
+    {context}
+
+    질문:
+    {question}
+    """
+
+    rag_prompt = ChatPromptTemplate.from_template(message)
+    # 2. chain
+    chain = {"context": retriever, "question": RunnablePassthrough()} | rag_prompt | watson_llm | StrOutputParser()
+
+    # 3. answer
+    answer = chain.invoke(question)
+    return answer
+
+async def rag_chat_stream(question: str):
+
+    faiss_store = FAISS.load_local("./db/vectorstore", watson_embedding, allow_dangerous_deserialization=True)
+
+    retriever = faiss_store.as_retriever(search_kwargs={"k": 3})
+
+    ### LLM
+    # 1. prompt
+    message = """\
+    당신은 PDF 기반 RAG AI 입니다.
+    다음 문서를 참고해서 질문에 답변하세요.
+
+    문서:
+    {context}
+
+    질문:
+    {question}
+    """
+
+    rag_prompt = ChatPromptTemplate.from_template(message)
+    # 2. chain
+    chain = {"context": retriever, "question": RunnablePassthrough()} | rag_prompt | watson_llm | StrOutputParser()
+
+    # 3. answer
+    # chain.stream() : 동기방식(요청 => 응답을 할 떄까지 기다리는 방식)
+    # chain.astream() : 비동기방식(다른 일 처리 가능)
+    async for chunk in chain.astream(question):
+        yield chunk
@@ -17,28 +17,58 @@ async function ask() {
 }

 // 파일 업로드
-document.querySelector("#uploadBtn").addEventListener("click", uploadFile)
-async function uploadFile()
-{
-    const fileInput = document.querySelector("#file");
+document.querySelector("#uploadBtn").addEventListener("click",uploadFile)
+async function uploadFile() {
+    const fileInput = document.querySelector("#file")
    // 첨부파일 정보 가져오기
-    const file = fileInput.files[0];
+    const file = fileInput.files[0]

-    if(!file)
-    {
-        alert("파일을 선택해주세요.");
+    if(!file){
+        alert('파일을 선택하세요');
        return;
    }

    // form 만들어 전송
-    const formData = new FormData()
-    formData.append("file", file)
+    const formData = new FormData();
+    formData.append("file",file);

-    const response = await fetch("/api/reg/question", {
-        method: "POST",
-        body: formData
+    const response = await fetch("/api/rag/upload",{
+        method:"POST",
+        body:formData
    })
    // 전송 후 answer 도착 시 answer 화면에 보여주기
    const answer = await response.json()
-    document.querySelector('#answer').textContent = answer.message
+    document.querySelector('#result').textContent = answer.message
+}
+
+document.querySelector("#askBtn").addEventListener("click", rag_ask)
+async function rag_ask() {
+    // 사용자가 질문 입력 시 질문을 서버로 전송
+    const question = document.querySelector('#question').value
+
+    const response = await fetch("/api/rag/question", {
+        method: "POST",
+        headers: {
+            "Content-Type": "application/json"
+        },
+        body: JSON.stringify({question : question})
+    })
+    // 전송 후 answer 도착 시 answer 화면에 보여주기
+    // const answer = await response.json()
+    // document.querySelector('#answer_result').textContent = answer.message
+
+    // stream 방식
+    const reader = response.body.getReader();
+    const decoder = new TextDecoder();
+    let answer = "";
+
+    while(true){
+        const { value, done } = await reader.read();
+
+        if (done) {
+            break;
+        }
+        answer += decoder.decode(value);
+        document.querySelector('#answer_result').textContent = answer;
+    }
 }
@@ -12,6 +12,12 @@
    </div>
    <hr>
    <div id="result"></div>
+    <hr>
+    <div>
+        <input type="text" name="question" id="question">
+        <button id="askBtn">질문</button>
+    </div>
+    <div id="answer_result"></div>
    <script src="{{ url_for('static', path='js/index.js')}}"></script>
 </body>
 </html>