# 1) 설치
# pip install llama-index llama-index-embeddings-openai llama-index-llms-openai
import os
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.embeddings.openai import OpenAIEmbedding
from llama_index.llms.openai import OpenAI
# 2) 모델 + 임베딩 구성
os.environ["OPENAI_API_KEY"] = "YOUR_KEY" # 또는 지원되는 LLM/임베딩 제공업체 사용
llm = OpenAI(model="gpt-4o-mini")
embed_model = OpenAIEmbedding(model="text-embedding-3-small")
# 3) 문서 로드 (예: ./data/*.pdf, .md, .txt)
docs = SimpleDirectoryReader("./data").load_data
# 4) 인덱스 구축
index = VectorStoreIndex.from_documents(docs, embed_model=embed_model)
# 5) 쿼리 엔진 생성 및 질문
query_engine = index.as_query_engine(llm=llm)
response = query_engine.query("문서에 언급된 주요 보안 관행은 무엇입니까?")
print(response)

from llama_index.core import SimpleDirectoryReader
from llama_index.readers.web import SimpleWebPageReader
file_docs = SimpleDirectoryReader("./policies").load_data
web_docs = SimpleWebPageReader(html_to_text=True).load_data
all_docs = file_docs + web_docs

from llama_index.core.node_parser import SentenceSplitter
from llama_index.core import Document
parser = SentenceSplitter(chunk_size=800, chunk_overlap=100)
nodes = []
for d in all_docs:
nodes.extend(parser.get_nodes_from_documents([Document(text=d.text, metadata=d.metadata)]))

from llama_index.core import VectorStoreIndex, SummaryIndex
from llama_index.core.retrievers import BM25Retriever
from llama_index.core.query_engine import RetrieverQueryEngine
# 미리 구문 분석된 노드의 벡터 인덱스
v_index = VectorStoreIndex(nodes)
# BM25 키워드 검색기
bm25_retriever = BM25Retriever.from_defaults(nodes=nodes, similarity_top_k=6)
# 하이브리드: 후보 병합, 재순위 지정
from llama_index.core.retrievers import RouterRetriever
from llama_index.retrievers.merge import MergerRetriever
v_retriever = v_index.as_retriever(similarity_top_k=6)
hybrid = MergerRetriever(retrievers=[v_retriever, bm25_retriever], top_k=8)
query_engine = RetrieverQueryEngine.from_args(retriever=hybrid)

from llama_index.postprocessor.flag_embedding_reranker import FlagEmbeddingReranker
from llama_index.core.query_engine import RetrieverQueryEngine
reranker = FlagEmbeddingReranker(top_n=5, model="BAAI/bge-reranker-base")
query_engine = v_index.as_query_engine(
similarity_top_k=12,
node_postprocessors=[reranker]
)

from llama_index.core.response_synthesizers import get_response_synthesizer
from llama_index.core import ServiceContext
synth = get_response_synthesizer(response_mode="tree_summarize")
query_engine = v_index.as_query_engine(response_synthesizer=synth)
ans = query_engine.query("온보딩 단계를 요약하고 소스를 인용하십시오.")
print(ans)

from llama_index.core.prompts import PromptTemplate
qa_tmpl = PromptTemplate(
"""
당신은 간결하고 증거 우선주의 도우미입니다. 제공된 컨텍스트만 사용하십시오.
확실하지 않으면 모른다고 말하십시오. answer, sources 키가 있는 JSON을 반환합니다.
질문: {query_str}
컨텍스트: {context_str}
"""
)
query_engine = v_index.as_query_engine(text_qa_template=qa_tmpl)

from llama_index.core.agent import ReActAgent
from llama_index.tools.sql import SQLQueryEngineTool
from sqlalchemy import create_engine
engine = create_engine("sqlite:///analytics.db")
sql_tool = SQLQueryEngineTool.from_engine(engine)
agent = ReActAgent.from_tools([sql_tool], llm=llm, verbose=True)
agent.chat("2025년 2분기 월별 이탈률은 얼마였습니까? 필요한 경우 DB를 쿼리하십시오.")

from llama_index.core.evaluation import FaithfulnessEvaluator, RelevancyEvaluator
faith = FaithfulnessEvaluator(llm=llm)
rel = RelevancyEvaluator(llm=llm)
pred = query_engine.query("정책에서 SOC 2 제어 체계를 나열하십시오.")
print("faithful?", faith.evaluate_response(pred))
print("relevant?", rel.evaluate_response(pred))

# 예: Chroma
# pip install chromadb llama-index-vector-stores-chroma
from llama_index.vector_stores.chroma import ChromaVectorStore
from llama_index.core import StorageContext
import chromadb
chroma_client = chromadb.PersistentClient(path="./chroma_store")
collection = chroma_client.get_or_create_collection("company_knowledge")
vector_store = ChromaVectorStore(chroma_collection=collection)
storage_context = StorageContext.from_defaults(vector_store=vector_store)
index = VectorStoreIndex.from_documents(all_docs, storage_context=storage_context)

# 예: 쿼리 시 메타데이터 기반 필터링
retriever = index.as_retriever(similarity_top_k=8)
retriever.metadata_filters = {"department": ["legal", "security"], "published": [True]}

# 최소 FastAPI 래퍼
# pip install fastapi uvicorn
from fastapi import FastAPI
app = FastAPI
qe = index.as_query_engine(llm=llm)
@app.post("/query")
async def query(payload: dict):
q = payload.get("q", "")
resp = qe.query(q)
return {"answer": str(resp), "sources": [s.node.metadata for s in resp.source_nodes]}