diff --git a/linebot_app.py b/linebot_app.py
index 0a7082a..b03835d 100755
--- a/linebot_app.py
+++ b/linebot_app.py
@@ -13,6 +13,7 @@ from langchain.vectorstores import Chroma
 from langchain.memory import ConversationBufferMemory
 from langchain.chains import RetrievalQA, ConversationalRetrievalChain
 from langchain.retrievers.multi_query import MultiQueryRetriever
+from langchain.prompts import PromptTemplate
 
 from flask import Flask, request, abort
 from linebot.v3 import (
@@ -44,15 +45,22 @@ log_file = open(log_file_name, "w")
 log_file.write("")
 log_file.close()
 
-
-
 # User Sessions
 user_sessions = {}
 
 # init vectorstore embedding
+qa_prompt_template = """
+Use the following pieces of context to answer the question at the end. 
+If you don't know the answer, just say that you don't know, don't try to make up an answer. 
+always add this exact keyphrase "answer_not_found" to the end of the text if you don't know the answer.
+{context}
+Question: {question}
+Helpful Answer:"""
+
 os.environ["OPENAI_API_KEY"] = srv_secrets.openai_key
 llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
 vectorstore = Chroma(persist_directory=srv_secrets.chroma_db_dir, embedding_function=OpenAIEmbeddings()) 
+qa_prompt = PromptTemplate.from_template()
 
 # Setup Logging
 logging.basicConfig()
@@ -109,7 +117,7 @@ def message_text(event):
     # User Session
     # create session if none exist
     if event.source.user_id not in user_sessions.keys():
-        user_sessions[event.source.user_id] = ConversationalRetrievalChain.from_llm(llm,retriever=retriever,memory=ConversationBufferMemory(memory_key="chat_history", return_messages=True))
+        user_sessions[event.source.user_id] = ConversationalRetrievalChain.from_llm(llm,retriever=retriever,memory=ConversationBufferMemory(memory_key="chat_history", return_messages=True, combine_docs_chain_kwargs={"prompt": qa_prompt}))
     
     # unique_docs = retriever_from_llm.get_relevant_documents(query=event.message.text)
     with ApiClient(configuration) as api_client: