.

about_kyobody.txt

@@ -0,0 +1,9 @@
+# ✍️ Kyobody 자기소개 🌼
+
+- **나이와 국적**: 1996년생 🇰🇷  
+- **직업**: 쿄바디 대표, 운동 건강 프로그램 크리에이터  
+- **학력**:
+  - **학사과정**: 국립 인천대학교 운동건강학부 2학년 2학기 재학 중  
+    (Division of Health and Kinesiology at INU)
+
+---

rag_gpt_app.py

@@ -12,7 +12,8 @@ from werkzeug.utils import safe_join
 from server.app import account, auth, bot_config, common, files, intervention, queries, sitemaps, urls
 from server.constant.constants import STATIC_DIR, MEDIA_DIR
 from server.logger.logger_config import my_logger as logger
-from server.app.queries import get_recall_documents
+from server.rag.pre_retrieval.query_transformation.rewrite import detect_query_lang
+from server.app.queries import get_recall_documents, refine_query
 
 from fastapi import FastAPI, HTTPException, Query
 
@@ -40,10 +41,18 @@ CORS(app)
 @fastapi_app.get("/api/get-context")
 async def get_context(
     query: str = Query(..., description="The search query for retrieving related documents"),
-    recall_top_k: int = 5,  # 기본값을 설정
-    min_relevance_score: float = 0.4  # 기본값을 설정
+    recall_top_k: int = 10,  # 기본값을 설정
+    min_relevance_score: float = 0.3  # 기본값을 설정
 ):
     try:
+        lang = detect_query_lang(query)
+        logger.warning(f"For query: '{query}', detect the language is '{lang}'!")
+
+        history_context = f"""Human: Hello
+        Assistant: I'm here to assist you with information related to `Kyobody (쿄바디 or 큐바디)`. If you have any specific questions about our services or need help, feel free to ask, and I'll do my best to provide you with accurate and relevant answers."""
+
+        query = refine_query(query, history_context, lang)
+
         # RECALL_TOP_K와 MIN_RELEVANCE_SCORE는 함수 인자로 받도록 설정
         results = get_recall_documents(
             query, 

server/constant/constants.py

@@ -26,7 +26,8 @@ ZHIPUAI_EMBEDDING_MODEL_NAME = "embedding-2"
 OLLAMA_EMBEDDING_MODEL_NAME = "mxbai-embed-large"
 
 # Maximum length of text chunks when splitting up large documents
-MAX_CHUNK_LENGTH = 1300
+# MAX_CHUNK_LENGTH = 1300
+MAX_CHUNK_LENGTH = 13000
 
 # Amount of overlap between consecutive text chunks
 CHUNK_OVERLAP = 100