39 lines
1.8 KiB
Python
39 lines
1.8 KiB
Python
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) |