added introductory prompt

app.py

@@ -12,11 +12,9 @@ DESCRIPTION = """\
 
 chat_interface =gr.ChatInterface(
         fn=handle_query, 
-        chatbot=gr.Chatbot(height=650),
+        chatbot=gr.Chatbot(height=600),
         textbox=gr.Textbox(placeholder="Chiedimi qualasiasi cosa relativa agli Osservatori", container=False, scale=7),
-        examples=[
-        ["Ciao, in cosa puoi aiutarmi?"],
-        ["Dimmi i risultati e le modalità di conduzione del censimento per favore"]]
+        #examples=[["Ciao, in cosa puoi aiutarmi?"],["Dimmi i risultati e le modalità di conduzione del censimento per favore"]]
         )
 
     

backend.py

@@ -13,7 +13,7 @@ from llama_cpp import Llama
 import spaces
 from huggingface_hub import login
 from llama_index.core.memory import ChatMemoryBuffer
-from typing import Iterator, List
+from typing import Iterator, List, Any
 from llama_index.core.chat_engine import CondensePlusContextChatEngine
 from llama_index.core.llms import ChatMessage, MessageRole
 
@@ -40,6 +40,17 @@ model.eval()
 Settings.embed_model = InstructorEmbedding(model_name="hkunlp/instructor-base")
 Settings.llm = GemmaLLMInterface()
 
+documents_paths = {
+    'blockchain': 'data/blockchainprova.txt',
+    'metaverse': 'data/metaverso',
+    'payment': 'data/payment'
+}
+
+session_state = {"documents_loaded": False, 
+                 "document_db": None, 
+                 "original_message": None, 
+                 "clarification": False}
+
 ############################---------------------------------
 
 # Get the parser
@@ -47,9 +58,9 @@ parser = SentenceSplitter.from_defaults(
                 chunk_size=256, chunk_overlap=64, paragraph_separator="\n\n"
             )
 
-def build_index():
+def build_index(path: str):
     # Load documents from a file
-    documents = SimpleDirectoryReader(input_files=["data/blockchainprova.txt"]).load_data()
+    documents = SimpleDirectoryReader(input_files=[path]).load_data()
     # Parse the documents into nodes
     nodes = parser.get_nodes_from_documents(documents)
     # Build the vector store index from the nodes
@@ -59,11 +70,39 @@ def build_index():
 
 
 @spaces.GPU(duration=20)
-def handle_query(query_str: str,
-    chat_history: list[tuple[str, str]],) -> Iterator[str]:
+def handle_query(query_str: str, 
+                 chat_history: list[tuple[str, str]], 
+                 session: dict[str, Any]) -> Iterator[str]:
+    
+    global index
     
-    index = build_index()
+    if not session["index"]:
+        matched_path = None
+        words = query_str.lower()
+        for key, path in documents_paths.items():
+            if key in words:
+                matched_path = path
+                break
+        if matched_path:
+            index = build_index(matched_path)
+            session["index"] = True
+            
+        else: ## CHIEDI CHIARIMENTO
+            conversation: List[ChatMessage] = []
+            for user, assistant in chat_history:
+                conversation.extend(
+                    [
+                        ChatMessage(role=MessageRole.USER, content=user),
+                        
+                        ChatMessage(role=MessageRole.ASSISTANT, content=assistant),
+                    ]
+                )
+            index = build_index("data/chiarimento.txt")     
     
+    else:
+        # The index is already built, no need to rebuild it.
+        pass    
+
     try:
           
         memory = ChatMemoryBuffer.from_defaults(token_limit=None)
@@ -82,7 +121,7 @@ def handle_query(query_str: str,
         chat_engine = index.as_chat_engine(
         chat_mode="condense_plus_context",
         memory=memory,
-        similarity_top_k=3, 
+        similarity_top_k=4, 
         response_mode="tree_summarize", #Good for summarization purposes
 
         context_prompt = (
@@ -92,9 +131,7 @@ def handle_query(query_str: str,
         "{context_str}"
         "\nIstruzione: Usa la cronologia delle chat precedenti, o il contesto sopra, per interagire e aiutare l'utente a rispondere alla sua domanda."
           ),
-        
         verbose=False,
-        
         )
         
         

data/chiarimento.txt

@@ -0,0 +1 @@
+In italiano, chiedi molto brevemente se la domanda si riferisce agli "Osservatori Blockchain", "Osservatori Payment" oppure "Osservatori Metaverse".