Se agrega al chat un componente spander que muestra las funestes de la respuesta para la citacion

app.py

@@ -6,6 +6,7 @@ from chats.streamlit_tools import import_file, clear_cache
 from streamlit_extras.add_vertical_space import add_vertical_space
 from langchain_tools.pdf_tools import PdfLangChain
 from langchain_tools.lc_tools import LangChainTools
+from chats.chat_tools import MessageManager
 
 
 # App title
@@ -121,12 +122,21 @@ if pdf_name:
                 input = "\n".join([msg["content"]
                                    for msg in st.session_state.messages])
 
-                query = qa.invoke({"question": f"{input}"},
+                query = qa.invoke({"question": f"{prompt}"},
                                   return_only_outputs=True)
 
-                response = query["answer"]
+                response_text = query["answer"]
+                documents_source = query["source_documents"]
+
+                messageManager = MessageManager()
+
+                citation: str = messageManager.generate_citations(
+                    documents_source)
+                # st.markdown(citation)
 
             with st.chat_message("assistant"):
-                st.write(response)
+                st.write(response_text)
                 st.session_state.messages.append(
-                    {"role": "assistant", "content": response})
+                    {"role": "assistant", "content": response_text})
+                expander = st.expander("Fuentes")
+                expander.markdown(citation)

chats/chat_tools.py

@@ -24,5 +24,17 @@ class MessageManager:
             print(f'{ia_emoticon} ' + Style.BRIGHT + Fore.YELLOW +
                   'IA:' + Style.RESET_ALL + f'{bot_response["answer"]}')
 
-    def generate_citations(self):
-        pass
+    def generate_citations(self, documents_source: list) -> str:
+
+        text_source: str = ""
+
+        for index, document in enumerate(documents_source):
+            quote: str = document.page_content
+            source: str = document.metadata['source'].replace(
+                'documents/pdfs/', '')
+            page: str = document.metadata['page'] + 1
+            fuente: str = f"**Fuente #{index + 1}:** \n '{quote}'\n(*{source}, P.{page})*"
+
+            text_source += fuente + "\n\n\n"
+
+        return text_source

langchain_tools/lc_tools.py

@@ -172,14 +172,17 @@ class LangChainTools():
 
         # Configura la memoria
         memory = ConversationBufferMemory(
-            memory_key="chat_history", return_messages=True)
+            memory_key="chat_history",
+            return_messages=True,
+            output_key='answer')
 
         # Define the Retrieval QA Chain to integrate the database and LLM
         conversation = ConversationalRetrievalChain.from_llm(
             _llm,
             retriever=_vectordb.as_retriever(),
             memory=memory,
-            verbose=False  # Modo verboso
+            verbose=False,  # Modo verboso
+            return_source_documents=True  # Devuelve los documentos fuente
         )
 
         return conversation