Update app.py

app.py

@@ -54,6 +54,7 @@ from geopy.geocoders import Nominatim
 from timezonefinder import TimezoneFinder
 from langchain_pinecone import Pinecone as PineconeLangchain
 import pinecone
+from uuid import uuid4
 
 pinecone.init(
     api_key=os.environ["PINECONE_API_KEY"],
@@ -235,23 +236,31 @@ session_file_hash = None
 session_retriever = None
 session_qa_chain = None
 
-def upload_and_chat(file, query):
-    global session_file_hash, session_retriever, session_qa_chain
+user_sessions = {}  # 用 dict 儲存每個 user_id 對應的 chain、retriever、hash
+
+def upload_and_chat(file, query, user_id=None):
+    if user_id is None:
+        user_id = str(uuid4())  # fallback
+
+    if user_id not in user_sessions:
+        user_sessions[user_id] = {
+            "file_hash": None,
+            "retriever": None,
+            "qa_chain": None
+        }
+
+    session = user_sessions[user_id]
 
     file_path = get_file_path(file)
     if file_path is None:
         return "Unable to obtain the uploaded file path."
     
-    # 計算目前上傳檔案的 hash 值
     import hashlib
     with open(file_path, "rb") as f:
         file_hash = hashlib.md5(f.read()).hexdigest()
 
-    # 如果是新文件 → 重建 retriever 和 chain
-    if file_hash != session_file_hash:
-        session_file_hash = file_hash
-
-        # Load and chunk the new document
+    if file_hash != session["file_hash"]:
+        session["file_hash"] = file_hash
         if file_path.lower().endswith(".pdf"):
             loader = PyPDFLoader(file_path)
         elif file_path.lower().endswith(".docx"):
@@ -261,18 +270,19 @@ def upload_and_chat(file, query):
         docs = loader.load()
         chunks = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50).split_documents(docs)
         pine_db = PineconeLangchain.from_documents(chunks, embeddings, index_name="rag-docs")
-        session_retriever = pine_db.as_retriever()
+        retriever = pine_db.as_retriever()
 
-        session_qa_chain = RetrievalQA.from_chain_type(
+        qa_chain = ConversationalRetrievalChain.from_llm(
             llm=llm_gpt4,
-            chain_type="stuff",
-            retriever=session_retriever,
-            return_source_documents=False,
-            chain_type_kwargs={"prompt": custom_prompt}
+            retriever=retriever,
+            memory=ConversationBufferMemory(memory_key="chat_history", return_messages=True),
+            combine_docs_chain_kwargs={"prompt": custom_prompt}
         )
 
-    # 用現有的 chain 執行 query
-    return session_qa_chain.run(query)
+        session["retriever"] = retriever
+        session["qa_chain"] = qa_chain
+
+    return session["qa_chain"].run(query)
     
 # tab 4 & 5 summary
 initial_prompt = PromptTemplate(