app.py

import streamlit as st
from PIL import Image
import textwrap
import google.generativeai as genai
import os
from PyPDF2 import PdfReader

# Function to display formatted Markdown text
def to_markdown(text):
    text = text.replace('•', '  *')
    return textwrap.indent(text, '> ', predicate=lambda _: True)

# Function to generate content using Gemini API
def generate_gemini_content(prompt, model_name='gemini-pro-vision', image=None):
    model = genai.GenerativeModel(model_name)
    if not image:
        st.warning("Por favor, agrega una imagen para usar el modelo gemini-pro-vision.")
        return None

    response = model.generate_content([prompt, image])
    return response

# Function to extract text from PDFs in a folder
def extract_text_from_pdf(pdf_path):
    text = ""
    with open(pdf_path, 'rb') as file:
        pdf_reader = PdfReader(file)
        num_pages = len(pdf_reader.pages)
        for page_num in range(num_pages):
            page = pdf_reader.pages[page_num]
            text += page.extract_text()
    return text

# Function to process PDFs in a folder
def process_pdfs_in_folder(folder_path):
    pdf_texts = {}
    for filename in os.listdir(folder_path):
        if filename.endswith(".pdf"):
            pdf_path = os.path.join(folder_path, filename)
            text = extract_text_from_pdf(pdf_path)
            pdf_texts[filename] = text
    return pdf_texts

# Function to get a response related to reference information
def get_reference_response():
    # Replace this with logic to generate a response based on reference information
    return "Lo siento, no puedo proporcionar una respuesta específica, pero aquí hay información de referencia relevante."

# Streamlit app
def main():
    st.set_page_config(page_title="MAX Chatbot - INIF", page_icon="🤖")

    # Configurar la API key de Gemini (reemplazar con tu clave de API de Gemini)
    genai.configure(api_key='AIzaSyA4k6JoFNZsf8L1ixLMMRjEMoBPns5SHZk')

    st.title("MAX Chatbot - INIF")
    st.sidebar.title("Configuración de MAX Chatbot")

    # Configurar la API key de INIF
    inif_api_key = 'AIzaSyA4k6JoFNZsf8L1ixLMMRjEMoBPns5SHZk'
    genai.configure(api_key=inif_api_key)

    # Seleccionar el modelo Gemini
    select_model = st.sidebar.selectbox("Selecciona el modelo", ["gemini-pro", "gemini-pro-vision"])

    # Inicializar la sesión de chat
    chat = genai.GenerativeModel(select_model).start_chat(history=[])

    # Definir función para obtener respuesta del modelo Gemini
    def get_response(messages):
        response = chat.send_message(messages, stream=True)
        return response

    # Historial del chat
    if "messages" not in st.session_state:
        st.session_state["messages"] = []

    messages = st.session_state["messages"]

    # Obtener texto de los PDFs en el directorio de ejecución del script
    current_directory = os.getcwd()
    pdf_texts = process_pdfs_in_folder(current_directory)

    # Incorporar el contenido de los PDFs al contexto del INIF
    inif_context = (
        "I am an informative data analyst chatbot named MAX, working for the National Institute of Fraud Research and Prevention (INIF), dedicated to fraud prevention and mitigation."
        " If you have questions related to fraud or prevention, feel free to ask. For inquiries about other topics, I'll redirect you to the fraud prevention context."
        "\n\nContact Information for INIF:"
        "\nPhone: +57 317 638 94 71"
        "\nEmail: atencionalcliente@inif.com.co"
        "\n\nOur Mission:"
        "\nTo be the most reliable engine of knowledge, research, and information in Colombia, capable of preventing and combating fraud through synergy between our team and companies."
        "\n\nOur Vision:"
        "\nTo lead the construction of a more honest culture, allowing us to progress as a society."
    )

    for pdf_name, pdf_text in pdf_texts.items():
        inif_context += f"\n\n**{pdf_name}**:\n{pdf_text}"

    # Mostrar mensajes del historial
    if messages:
        for message in messages:
            role, parts = message.values()
            if role.lower() == "user":
                st.markdown(f"Tú: {parts[0]}")
            elif role.lower() == "model":
                st.markdown(f"Assistant: {to_markdown(parts[0])}")

    # Entrada del usuario
    user_input = st.text_area("Tú:")

    # Concatenar el contexto del INIF al input del usuario
    user_input_with_context = f"{user_input}\n\n{inif_context}"

    # Get optional image input if the model selected is 'gemini-pro-vision'
    image_file = None
    if select_model == 'gemini-pro-vision':
        image_file = st.file_uploader("Sube una imagen (si aplica):", type=["jpg", "jpeg", "png"])

        # Display image if provided
        if image_file:
            st.image(image_file, caption="Imagen subida", use_column_width=True)

    # Botón para enviar mensaje o generar contenido según el modelo seleccionado
    if st.button("Enviar / Generar Contenido"):
        if user_input:
            messages.append({"role": "user", "parts": [user_input]})
            if select_model == 'gemini-pro-vision':
                # Modelo Gemini Vision Pro seleccionado
                if not image_file:
                    st.warning("Por favor, proporciona una imagen para el modelo gemini-pro-vision.")
                else:
                    image = Image.open(image_file)
                    response = generate_gemini_content(user_input_with_context, model_name=select_model, image=image)
                    if response:
                        if response.candidates:
                            parts = response.candidates[0].content.parts
                            generated_text = parts[0].text if parts else get_reference_response()
                            st.markdown(f"Assistant: {to_markdown(generated_text)}")
                            messages.append({"role": "model", "parts": [generated_text]})
                        else:
                            st.markdown(f"Assistant: {to_markdown(get_reference_response())}")
                            messages.append({"role": "model", "parts": [get_reference_response()]})
            else:
                # Otros modelos Gemini seleccionados
                response = get_response(user_input_with_context)

                # Mostrar respuesta del modelo solo una vez
                res_text = ""
                for chunk in response:
                    res_text += chunk.text
                if not res_text.strip():
                    st.markdown(f"Assistant: {to_markdown(get_reference_response())}")
                    messages.append({"role": "model", "parts": [get_reference_response()]})
                else:
                    st.markdown(f"Assistant: {to_markdown(res_text)}")
                    messages.append({"role": "model", "parts": [res_text]})

    # Actualizar historial de mensajes en la sesión de Streamlit
    st.session_state["messages"] = messages

if __name__ == "__main__":
    main()