Update app.py

app.py

@@ -91,7 +91,22 @@ llm_chain = create_stuff_documents_chain(llm=llm, prompt=qa_prompt)
 
 #rag_chain = create_retrieval_chain(history_aware_retriever, llm_chain)
 
-def truncate_history(chat_history, max_chars=4000):
+# Preload retrievers once
+retrievers = {
+    "en": FAISS.load_local(
+        folder_path=INDEX_PATHS["en"],
+        embeddings=embedding,
+        allow_dangerous_deserialization=True
+    ).as_retriever(search_kwargs={"k": 2}),
+    "fr": FAISS.load_local(
+        folder_path=INDEX_PATHS["fr"],
+        embeddings=embedding,
+        allow_dangerous_deserialization=True
+    ).as_retriever(search_kwargs={"k": 2}),
+}
+
+# Truncate long history
+def truncate_history(chat_history, max_chars=1500):
     total = 0
     trimmed = []
     for q, a in reversed(chat_history):
@@ -102,46 +117,31 @@ def truncate_history(chat_history, max_chars=4000):
         total += pair_len
     return trimmed
 
+# Simpler, faster RAG function
 def rag_tool_func(input_question: str, chat_history: list = None) -> str:
-    # Detect language
     lang = detect(input_question)
-    lang = "fr" if lang == "fr" else "en"  # fallback to English if unsure
+    lang = "fr" if lang == "fr" else "en"
 
-    # Load correct vector store
-    retriever = FAISS.load_local(
-        folder_path=INDEX_PATHS[lang],
-        embeddings=embedding,
-        allow_dangerous_deserialization=True
-    ).as_retriever(search_kwargs={"k": 2})
-
-    # Create history-aware retriever
-    history_aware_retriever = create_history_aware_retriever(
-        llm, retriever, condense_question_prompt
-    )
-
-    # Create LLM QA chain
-    llm_chain = create_stuff_documents_chain(llm=llm, prompt=qa_prompt)
-    rag_chain = create_retrieval_chain(history_aware_retriever, llm_chain)
-
-    # Format chat history
+    retriever = retrievers[lang]
 
+    # Format chat history (optional, for prompt context)
     chat_history = truncate_history(chat_history)
-    
-    messages = []
+    history_str = ""
     if isinstance(chat_history, list):
         for q, a in chat_history:
-            messages.append(("human", q))
-            messages.append(("ai", a))
+            history_str += f"User: {q}\nAssistant: {a}\n"
 
+    rag_chain = create_retrieval_chain(retriever, create_stuff_documents_chain(llm=llm, prompt=qa_prompt))
     result = rag_chain.invoke({
         "input": input_question,
-        "chat_history": messages
+        "chat_history": history_str
     })
     return result["answer"]
 
 
 
 
+
 chat_history = []  # Global chat history
 
 def chatbot_interface(user_input, history):