app.py

import streamlit as st
import os
from langchain.document_loaders import PyPDFDirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings
from langchain.llms import OpenAI
from langchain_core.output_parsers import StrOutputParser
from langchain import hub
from langchain_core.runnables import RunnablePassthrough
from langchain.indexes import VectorstoreIndexCreator

# Definir la clase Document
class Document:
    def __init__(self, page_content, metadata=None):
        self.page_content = page_content
        self.metadata = metadata if metadata else {}

# Configuración inicial y variables de entorno
st.title("Busca en PDFs")

# Entradas de usuario para las API keys
st.write("Obten tu api-key de https://smith.langchain.com")
OPENAI_API_KEY = st.text_input('OpenAI Api Key', type='password')
SMITH_APIKEY = st.text_input('Smith Api Key', type='password')

if OPENAI_API_KEY and SMITH_APIKEY:
    os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY
    os.environ["LANGCHAIN_TRACING_V2"] = "true"
    os.environ["LANGCHAIN_API_KEY"] = SMITH_APIKEY

    # Función principal
    def main():
        st.write("Aplicación en Streamlit ejecutándose correctamente")

        # Configura las carpetas necesarias
        source_data_folder = "MisDatos"
        path_db = "VectorDB"
        os.makedirs(source_data_folder, exist_ok=True)
        os.makedirs(path_db, exist_ok=True)

        # Carga de archivos PDF
        uploaded_files = st.file_uploader("Sube archivos PDF", type="pdf", accept_multiple_files=True)
        if uploaded_files:
            for uploaded_file in uploaded_files:
                with open(os.path.join(source_data_folder, uploaded_file.name), "wb") as f:
                    f.write(uploaded_file.getbuffer())
            st.write(f"{len(uploaded_files)} archivos subidos exitosamente")

            # Leer y procesar PDFs
            loader = PyPDFDirectoryLoader(source_data_folder)
            data_on_pdf = loader.load()
            st.write(f"Se han cargado {len(data_on_pdf)} documentos")

            # Preprocesar texto
            def preprocess_text(text):
                cleaned_text = ' '.join(text.split())
                return cleaned_text

            # Particionar datos
            text_splitter = RecursiveCharacterTextSplitter(
                separators=["\n\n", "\n", ". ", " ", ""],
                chunk_size=1024,
                chunk_overlap=200
            )
            splits = []
            for doc in data_on_pdf:
                cleaned_text = preprocess_text(doc.page_content)
                split_docs = text_splitter.split_documents([Document(page_content=cleaned_text, metadata=doc.metadata)])
                splits.extend(split_docs)
            st.write(f"Se han creado {len(splits)} fragmentos")

            # Embeddings
            embeddings_model = OpenAIEmbeddings(model="text-embedding-ada-002")

            # Crear y persistir la base de datos vectorial
            vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings_model, persist_directory=path_db)

            # Crear el índice usando VectorstoreIndexCreator
            index_creator = VectorstoreIndexCreator(embedding=embeddings_model)
            index = vectorstore.as_retriever()
            st.write(f"El tipo de índice es: {type(index)}")

            # Configuración del LLM
            llm = OpenAI(model_name="gpt-3.5-turbo-0125", temperature=0.8)

            # Configuración del retriever
            retriever = vectorstore.as_retriever()

            # Cargar el prompt desde LangChain Hub
            prompt = hub.pull("rlm/rag-prompt")

            def format_docs(docs):
                return "\n\n".join(doc.page_content for doc in docs)

            # Definición del pipeline de procesamiento
            rag_chain = (
                {"context": retriever | format_docs, "question": RunnablePassthrough()}
                | prompt
                | llm
                | StrOutputParser()
            )

            # Interacción con el usuario
            pregunta = st.text_area("Haz una pregunta sobre los documentos:")
            if pregunta:
                try:
                    response = rag_chain.invoke(pregunta)
                    st.markdown(response)
                except Exception as e:
                    st.error(f"Error al procesar la pregunta: {e}")

    if __name__ == "__main__":
        main()
else:
    st.write("Por favor, proporciona las API keys para continuar.")