Update app.py

app.py

@@ -10,15 +10,20 @@ from huggingface_hub import InferenceClient
 from rerankers import Reranker
 import os
 
-loader = PyPDFLoader("Constitucion_española.pdf")
-documents = loader.load()
+embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
+persist_directory = "db"
 
-text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
-docs_split = text_splitter.split_documents(documents)
+if not os.path.exists(persist_directory):
+    loader = PyPDFLoader("Constitucion_española.pdf")
+    documents = loader.load()
 
-embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
+    docs_split = text_splitter.split_documents(documents)
 
-vectordb = Chroma.from_documents(docs_split, embedding_function)
+    vectordb = Chroma.from_documents(docs_split, embedding_function, persist_directory=persist_directory)
+    vectordb.persist()
+else:
+    vectordb = Chroma(persist_directory=persist_directory, embedding_function=embedding_function)
 
 client = InferenceClient("google/flan-t5-base", token=os.getenv("HUGGINGFACEHUB_API_TOKEN"))
 
@@ -51,14 +56,12 @@ def test_rag_reranking(query, ranker):
 
     print(f"📚 Contextos pasados al ranker: {len(context)}")
 
-    # ✅ Corregido: pasar solo lista de strings
     context_strings = [str(c) for c in context]
     #print(help(Reranker.models.ColBERTRanker.rank))
     reranked = ranker.rank(query=query, docs=context_strings)
     
     print(f"🏅 Resultado del reranker: {reranked}")
 
-    # ✅ Seguridad en el acceso al mejor contexto
     best_context = reranked[0].document.text  
     print(f"🧠 Contexto elegido: {best_context[:500]}...")
 
@@ -67,8 +70,6 @@ def test_rag_reranking(query, ranker):
 
     return respuesta
 
-
-
 def responder_chat(message, history):
     respuesta = test_rag_reranking(message, ranker)
     return respuesta