Fix vector DB loading: robust unzip + no re-embedding

src/chatbot.py

@@ -9,17 +9,33 @@ import sys
 from pathlib import Path
 from typing import List, Dict
 
+# 프로젝트 루트 설정
 PROJECT_ROOT = Path(__file__).parent.parent
 sys.path.insert(0, str(PROJECT_ROOT / "src"))
 
+# =========================
+# imports
+# =========================
 from retrieval import FaissRetriever
-from embedding import EmbeddingModel
+from embedding import embed_query
 from llmmodel import get_llm
 from prompt import (
     create_rag_prompt,
     format_answer
 )
 
+# =========================
+# 최소 래퍼: EmbeddingModel
+# =========================
+class EmbeddingModel:
+    """
+    FAISS 검색용 쿼리 임베딩 래퍼
+    - vector DB는 사전 구축됨
+    - encode(query)만 제공
+    """
+    def encode(self, text: str):
+        return embed_query(text)
+
 
 class SocialCultureChatbot:
     """
@@ -38,7 +54,7 @@ class SocialCultureChatbot:
         print("=" * 60)
 
         # 임베딩 모델 (encode only)
-        print("\n🔤 임베딩 모델 로드 중...")
+        print("\n🔤 임베딩 래퍼 초기화 중...")
         self.embedding_model = EmbeddingModel()
 
         # FAISS 검색기
@@ -60,7 +76,7 @@ class SocialCultureChatbot:
         """
         질문에 답변합니다.
         """
-        # 1. 검색
+        # 1. FAISS 검색
         results = self.retriever.retrieve(
             query=question,
             top_k=self.top_k
@@ -71,7 +87,7 @@ class SocialCultureChatbot:
             [f"[{i+1}] {r['text']}" for i, r in enumerate(results)]
         )
 
-        # 3. 프롬프트
+        # 3. 프롬프트 생성
         prompt = create_rag_prompt(question, context)
 
         # 4. LLM 호출

src/retrieval.py

@@ -5,24 +5,39 @@
 - ❌ 문서 재임베딩 없음
 """
 
-from typing import List, Dict, Tuple
+from typing import List, Dict
 from pathlib import Path
+import numpy as np
 import sys
 
 PROJECT_ROOT = Path(__file__).parent.parent
 sys.path.insert(0, str(PROJECT_ROOT / "src"))
 
-from search_model_setup import get_search_model
-from embedding import embed_text
+import faiss
+import json
 
 
-class Retriever:
+class FaissRetriever:
     """
     FAISS 기반 RAG 검색기
     """
 
-    def __init__(self, collection_name: str = "social_culture"):
-        self.search_model = get_search_model(collection_name)
+    def __init__(
+        self,
+        vector_db_dir: str,
+        embedding_model
+    ):
+        self.vector_db_dir = Path(vector_db_dir)
+        self.embedding_model = embedding_model
+
+        # FAISS index
+        index_path = self.vector_db_dir / "faiss_index.bin"
+        self.index = faiss.read_index(str(index_path))
+
+        # metadata
+        meta_path = self.vector_db_dir / "embeddings_metadata.json"
+        with open(meta_path, "r", encoding="utf-8") as f:
+            self.metadata = json.load(f)
 
     def retrieve(
         self,
@@ -30,34 +45,27 @@ class Retriever:
         top_k: int = 5
     ) -> List[Dict]:
         """
-        FAISS 검색만 수행
+        FAISS 검색 수행
         """
-        return self.search_model.search(query, top_k=top_k)
-
-    def retrieve_with_context(
-        self,
-        query: str,
-        top_k: int = 5
-    ) -> Tuple[List[Dict], str]:
-        """
-        검색 결과 + context 문자열 반환
-        """
-        results = self.retrieve(query, top_k=top_k)
-
-        context_parts = []
-        for i, r in enumerate(results):
-            page = r["metadata"].get("page_number", "N/A")
-            context_parts.append(
-                f"[문서 {i+1}] (페이지 {page})\n{r['text']}"
-            )
-
-        context = "\n\n---\n\n".join(context_parts)
-        return results, context
+        # 1. query embedding
+        query_vec = self.embedding_model.encode(query)
+        query_vec = np.array([query_vec]).astype("float32")
 
+        # 2. FAISS search
+        distances, indices = self.index.search(query_vec, top_k)
 
-def get_retriever(collection_name: str = "social_culture") -> Retriever:
-    return Retriever(collection_name)
+        # 3. 결과 구성
+        results = []
+        for rank, idx in enumerate(indices[0]):
+            if idx < 0:
+                continue
 
-# alias for backward compatibility
-FaissRetriever = Retriever
+            meta = self.metadata[idx]
+            results.append({
+                "rank": rank + 1,
+                "score": float(distances[0][rank]),
+                "text": meta["text"],
+                "metadata": meta.get("metadata", {})
+            })
 
+        return results