callcenter 프로젝트 완료

- 상담 기록 분석 및 요약
- 상담 기록 db 저장
- 상담 평가
- 상담 평가 db 저장

project/CALLCENTER_APP/backend/main.py

@@ -3,6 +3,9 @@ from starlette.staticfiles import StaticFiles
 from contextlib import asynccontextmanager
 
 from backend.repository.db_init import Base, SessionLocal, engine
+
+from backend.routers.evalu_router import router as evalu_router
+from backend.routers.assistant_router import router as assistant_router
 from backend.routers.call_router import router as call_router
 from backend.repository.seed import seed_customers
 
@@ -31,4 +34,6 @@ app = FastAPI(title="상담 LLM", version="1.0", lifespan=lifespan)
 app.mount("/static", StaticFiles(directory="backend/static"), name="static")
 
 # 라우터 등록
-app.include_router(call_router)
+app.include_router(call_router)
+app.include_router(assistant_router)
+app.include_router(evalu_router)

project/CALLCENTER_APP/backend/prompts/all_prompt.py

@@ -25,4 +25,48 @@ SUMMARY_SYSTEM_PROMPT = """
 
 7. resolution
     - 상담사가 제공한 해결 방법
+"""
+
+CALL_ASSISTANT_PROMPT = """
+당신은 콜센터 상담 지원 ai 입니다.
+
+제공된 정보만 활용하여 답변하세요.
+참고 정보가 부족하면 추측하지 말고 추가 확인이 필요하다고 답변하세요.
+
+[지식문서]
+{sim_context}
+
+[고객 상담 이력]
+{customer_text}
+
+[질문]
+{question}
+"""
+
+CALL_EVALUATION_PROMPT = """
+당신은 콜센터 QA 평가 전문가입니다.
+
+다음 항목을 평가하세요.
+1. 고객 신원 확인
+2. 공감 표현
+3. 문제 해결
+4. 설문 조사 안내
+
+각 항목에 대해
+- True / False
+- 판단 근거(reason)
+
+를 제공하세요.
+
+특히 공감 표현은 아래와 같이 명시적 표현이 있을 때만 True로 판단하세요.
+
+예시:
+- 불편을 드려 죄송합니다.
+- 많이 답답하셨겠습니다.
+- 불편을 겪으셔서 죄송합니다.
+
+단순히 문제 해결 절차를 안내하는 경우는 공감 표현으로 간주하지 마세요.
+
+상담기록:
+{transcript}
 """

project/CALLCENTER_APP/backend/repository/models.py

@@ -27,4 +27,27 @@ class CallHistory(Base):
     customer_issue:Mapped[str]
     resolution:Mapped[str]
     # created_at:Mapped[datetime] = mapped_column(server_default=func.now(), nullable=False)
-    created_at:Mapped[datetime] = mapped_column(default=datetime.now, nullable=False)
+    created_at:Mapped[datetime] = mapped_column(default=datetime.now, nullable=False)
+
+# 상담 평가 저장
+class CallEvaluation(Base):
+    __tablename__ = 'call_evaluation'
+
+    evaluation_id:  Mapped[int] = mapped_column(primary_key=True, autoincrement=True)
+    call_id: Mapped[int] = mapped_column(ForeignKey('call_history.call_id'))
+
+    # 평가 기준
+    # 고객 신원 확인
+    identity_verification:Mapped[bool] = mapped_column(default=False)
+    identity_verification_reason:Mapped[str]
+    # 공감 표현
+    empathy:Mapped[bool] = mapped_column(default=False)
+    empathy_reason:Mapped[str]
+    # 문제 해결
+    issue_resolution:Mapped[bool] = mapped_column(default=False)
+    issue_resolution_reason:Mapped[str]
+    # 설문 조사 안내
+    survey_guidance:Mapped[bool] = mapped_column(default=False)
+    survey_guidance_reason:Mapped[str]
+    score:Mapped[int]
+    created_at:Mapped[datetime] = mapped_column(default=datetime.now, nullable=False)

project/CALLCENTER_APP/backend/routers/assistant_router.py

@@ -0,0 +1,14 @@
+from fastapi import APIRouter, Depends
+from backend.schemas.assistant_schema import AssistantResponse, AssistantRequest
+from backend.schemas.summary_schema import SummaryRequest, CallSummary, CallRequest
+from backend.services.assistant_service import answer_assistant_question
+from backend.services.call_service import summary_call, create_call_history
+from sqlalchemy.orm import Session
+from backend.repository.db_init import get_db
+
+
+router = APIRouter(prefix="/api/assistant", tags=["Assistant"])
+
+@router.post("", response_model=AssistantResponse)
+def ask_assistant(req:AssistantRequest, db:Session = Depends(get_db)):
+    return answer_assistant_question(customer_id = req.customer_id, question = req.question, db = db)

project/CALLCENTER_APP/backend/routers/call_router.py

@@ -13,6 +13,5 @@ def generate_summary(req:SummaryRequest):
     return summary_call(req.transcript)
 
 @router.post("/save", response_model=CallSummary)
-# @router.post("/save", response_model=CallRequest)
 def create_call(req: CallRequest, db:Session = Depends(get_db)):
     return create_call_history(req, db)

project/CALLCENTER_APP/backend/routers/evalu_router.py

@@ -0,0 +1,11 @@
+from fastapi import APIRouter, Depends
+from backend.schemas.evaluation_schema import EvaluationResponse
+from sqlalchemy.orm import Session
+from backend.repository.db_init import get_db
+from backend.services.evalu_service import get_call_evaluation
+
+router = APIRouter(prefix="/api/evaluation", tags=["evaluation"])
+
+@router.get("/{call_id}", response_model=EvaluationResponse)
+def read_evaluation(call_id : int, db:Session = Depends(get_db)):
+    return get_call_evaluation(call_id = call_id, db = db)

project/CALLCENTER_APP/backend/schemas/assistant_schema.py

@@ -0,0 +1,8 @@
+from pydantic import BaseModel
+
+class AssistantRequest(BaseModel):
+    customer_id: int
+    question: str
+
+class AssistantResponse(BaseModel):
+    answer: str

project/CALLCENTER_APP/backend/schemas/evaluation_schema.py

@@ -0,0 +1,14 @@
+from pydantic import BaseModel
+from datetime import datetime
+
+class EvaluationResponse(BaseModel):
+    identity_verification:bool
+    identity_verification_reason:str
+    empathy:bool
+    empathy_reason:str
+    issue_resolution:bool
+    issue_resolution_reason:str
+    survey_guidance:bool
+    survey_guidance_reason:str
+    score: int
+    created_at: datetime

project/CALLCENTER_APP/backend/schemas/summary_schema.py

@@ -25,4 +25,15 @@ class CallCreate(BaseModel):
     category: str
     sentiment: str
     customer_issue: str
-    resolution: str
+    resolution: str
+
+# 평가 스키마
+class CallEvaluationResponse(BaseModel):
+    identity_verification:bool
+    identity_verification_reason:str
+    empathy:bool
+    empathy_reason:str
+    issue_resolution:bool
+    issue_resolution_reason:str
+    survey_guidance:bool
+    survey_guidance_reason:str

project/CALLCENTER_APP/backend/scripts/build_vector_db.py

@@ -1,4 +1,4 @@
-from langchain_community.vectorstores import Chroma
+from langchain_chroma import Chroma
 from backend.ai.embedding import watson_embedding
 from langchain_core.documents import Document
 from langchain_text_splitters import RecursiveCharacterTextSplitter

project/CALLCENTER_APP/backend/services/assistant_service.py

@@ -0,0 +1,39 @@
+from backend.ai.llm import hugging_llm
+from langchain_core.prompts import ChatPromptTemplate
+from backend.prompts.all_prompt import SUMMARY_SYSTEM_PROMPT, CALL_ASSISTANT_PROMPT
+from backend.repository.models import CallHistory
+from backend.schemas.summary_schema import CallSummary, CallCreate
+from sqlalchemy.orm import Session
+from backend.schemas.assistant_schema import AssistantRequest
+from langchain_chroma import Chroma
+from backend.ai.embedding import watson_embedding
+from langchain_core.output_parsers import StrOutputParser
+
+# 질의 응답
+def answer_assistant_question(customer_id:int, question:str, db:Session):
+    # 1 단계 : 벡터 DB에서 질의
+    # 벡터db 불러오기
+    vectorstore = Chroma(embedding_function=watson_embedding, persist_directory="./vectordb")
+    # as_retriever()
+    retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
+    # invoke() = docs => page_content join
+    docs = retriever.invoke(question)
+    sim_context = "\n\n".join(doc.page_content for doc in docs)
+    # 2 단계 : DB 검색
+    # 고객이 이전에 질문한 내역을 추출
+    if customer_id:
+        histories = db.query(CallHistory).filter(CallHistory.customer_id == customer_id).order_by(CallHistory.created_at.desc()).limit(5).all()
+
+        # 문제, 해결 컬럼만 문자열로 추출
+        customer_text ="\n".join([f"""
+        문제: {h.customer_issue}\n
+        해결: {h.resolution}
+""" for h in histories])
+
+    # 1, 2 단계 => LLM => 답변 생성
+    prompt = ChatPromptTemplate.from_template(CALL_ASSISTANT_PROMPT)
+    chain = prompt | hugging_llm | StrOutputParser()
+    result = chain.invoke({"sim_context": sim_context, "customer_text": customer_text, "question": question})
+
+    return {"answer" : result}
+    # return AssistantRequest(answer=result)

project/CALLCENTER_APP/backend/services/call_service.py

@@ -1,8 +1,8 @@
 from backend.ai.llm import hugging_llm
 from langchain_core.prompts import ChatPromptTemplate
-from backend.prompts.all_prompt import SUMMARY_SYSTEM_PROMPT
-from backend.repository.models import CallHistory
-from backend.schemas.summary_schema import CallSummary, CallCreate
+from backend.prompts.all_prompt import SUMMARY_SYSTEM_PROMPT, CALL_EVALUATION_PROMPT
+from backend.repository.models import CallHistory, CallEvaluation
+from backend.schemas.summary_schema import CallSummary, CallCreate, CallEvaluationResponse
 from sqlalchemy.orm import Session
 
 from backend.schemas.summary_schema import CallRequest
@@ -41,15 +41,51 @@ def save_call_history(db, data):
 
     return history
 
-def evaluate_call():
+def evaluate_call(transcript:str):
     """상담  평가"""
+    structured_llm = hugging_llm.with_structured_output(CallEvaluationResponse)
+    prompt = ChatPromptTemplate.from_template(CALL_EVALUATION_PROMPT)
 
-    pass
+    chain = prompt | structured_llm
+    return chain.invoke({"transcript":transcript})
 
-def save_call_evaluation():
+def calculate_score(evaluation):
+    score = 0
+
+    if evaluation.identity_verification:
+        score += 25
+
+    if evaluation.empathy:
+        score += 25
+
+    if evaluation.issue_resolution:
+        score += 25
+
+    if evaluation.survey_guidance:
+        score += 25
+
+    return score
+
+def save_call_evaluation(db:Session, call_id:int, evaluation:CallEvaluationResponse):
     """상담 평가 내용 저장"""
+    score = calculate_score(evaluation)
+
+    entity = CallEvaluation(
+        call_id = call_id,
+        identity_verification = evaluation.identity_verification,
+        identity_verification_reason = evaluation.identity_verification_reason,
+        empathy = evaluation.empathy,
+        empathy_reason = evaluation.empathy_reason,
+        issue_resolution = evaluation.issue_resolution,
+        issue_resolution_reason = evaluation.issue_resolution_reason,
+        survey_guidance = evaluation.survey_guidance,
+        survey_guidance_reason = evaluation.survey_guidance_reason,
+        score = score,
+    )
+    db.add(entity)
+    db.commit()
+    db.refresh(entity)
 
-    pass
 
 def create_call_history(req: CallRequest, db:Session):
     # 요약
@@ -66,7 +102,19 @@ def create_call_history(req: CallRequest, db:Session):
         customer_issue = summary.customer_issue,
         resolution = summary.resolution
     )
+    history = save_call_history(db = db, data = call_data)
 
-    return save_call_history(db = db, data = call_data)
-
+    # 상담 평가
+    evaluation = evaluate_call(req.transcript)
+    # 평가 저장
+    save_call_evaluation(db = db, call_id=history.call_id, evaluation=evaluation)
 
+    return CallSummary(
+        summary=summary.summary,
+        keywords=summary.keywords,
+        category=summary.category,
+        sentiment=summary.sentiment,
+        action_items=summary.action_items,
+        customer_issue=summary.customer_issue,
+        resolution=summary.resolution,
+    )

project/CALLCENTER_APP/backend/services/evalu_service.py

@@ -0,0 +1,23 @@
+from sqlalchemy.orm import Session
+from backend.repository.models import CallEvaluation
+from backend.schemas.evaluation_schema import EvaluationResponse
+
+
+# 질의 응답
+def get_call_evaluation(call_id:int, db:Session):
+    # DB 검색 - call_id와 일치하는 정보 추출
+    if call_id:
+        evaluation = db.query(CallEvaluation).filter(CallEvaluation.call_id == call_id).first()
+
+    return EvaluationResponse(
+        identity_verification=evaluation.identity_verification,
+        identity_verification_reason=evaluation.identity_verification_reason,
+        empathy=evaluation.empathy,
+        empathy_reason=evaluation.empathy_reason,
+        issue_resolution=evaluation.issue_resolution,
+        issue_resolution_reason=evaluation.issue_resolution_reason,
+        survey_guidance=evaluation.survey_guidance,
+        survey_guidance_reason=evaluation.survey_guidance_reason,
+        score=evaluation.score,
+        created_at=evaluation.created_at,
+    )