app.py

import streamlit as st
from PIL import Image
import random
import time
import streamlit_analytics
from dotenv import load_dotenv
import pickle
from huggingface_hub import Repository
from PyPDF2 import PdfReader
from streamlit_extras.add_vertical_space import add_vertical_space
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import FAISS
from langchain.llms import OpenAI
from langchain.chains.question_answering import load_qa_chain
from langchain.callbacks import get_openai_callback
import os

import pandas as pd
import pydeck as pdk
from urllib.error import URLError

# Initialize session state variables
if 'chat_history_page1' not in st.session_state:
    st.session_state['chat_history_page1'] = []

if 'chat_history_page2' not in st.session_state:
    st.session_state['chat_history_page2'] = []

if 'chat_history_page3' not in st.session_state:
    st.session_state['chat_history_page3'] = []



# Step 1: Clone the Dataset Repository
repo = Repository(
    local_dir="Private_Book",  # Local directory to clone the repository
    repo_type="dataset",  # Specify that this is a dataset repository
    clone_from="Anne31415/Private_Book",  # Replace with your repository URL
    token=os.environ["HUB_TOKEN"]  # Use the secret token to authenticate
)
repo.git_pull()  # Pull the latest changes (if any)

# Step 2: Load the PDF File
pdf_path = "Private_Book/KH_Reform_230124.pdf"  # Replace with your PDF file path

pdf_path2 = "Private_Book/Buch_23012024.pdf"  

pdf_path3 = "Private_Book/Kosten_Grunddaten_KH_230124.pdf"  

api_key = os.getenv("OPENAI_API_KEY")
# Retrieve the API key from st.secrets



# Updated load_vector_store function with Streamlit text outputs and directory handling for Git
@st.cache_data(persist="disk")
def load_vector_store(file_path, store_name, force_reload=False):
    local_repo_path = "Private_Book"
    vector_store_path = os.path.join(local_repo_path, f"{store_name}.pkl")

    # Check if vector store already exists and force_reload is False
    if not force_reload and os.path.exists(vector_store_path):
        with open(vector_store_path, "rb") as f:
            VectorStore = pickle.load(f)
        #st.text(f"Loaded existing vector store from {vector_store_path}")
    else:
        # Load and process the PDF, then create the vector store
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200, length_function=len)
        text = load_pdf_text(file_path)
        chunks = text_splitter.split_text(text=text)
        embeddings = OpenAIEmbeddings()
        VectorStore = FAISS.from_texts(chunks, embedding=embeddings)

        # Serialize the vector store
        with open(vector_store_path, "wb") as f:
            pickle.dump(VectorStore, f)
        #st.text(f"Created and saved vector store at {vector_store_path}")

        # Change working directory for Git operations
        original_dir = os.getcwd()
        os.chdir(local_repo_path)
        
        try:
            # Check current working directory and list files for debugging
            #st.text(f"Current working directory: {os.getcwd()}")
            #st.text(f"Files in current directory: {os.listdir()}")
        
            # Adjusted file path for Git command
            repo.git_add(f"{store_name}.pkl")  # Use just the file name
            repo.git_commit(f"Update vector store: {store_name}")
            repo.git_push()
            #st.text("Committed and pushed vector store to repository.")
        except Exception as e:
            st.error(f"Error during Git operations: {e}")
        finally:
            # Change back to the original directory
            os.chdir(original_dir)

    return VectorStore


# Utility function to load text from a PDF
def load_pdf_text(file_path):
    pdf_reader = PdfReader(file_path)
    text = ""
    for page in pdf_reader.pages:
        text += page.extract_text() or ""  # Add fallback for pages where text extraction fails
    return text

def load_chatbot():
    #return load_qa_chain(llm=OpenAI(), chain_type="stuff")
    return load_qa_chain(llm=OpenAI(model_name="gpt-3.5-turbo-instruct"), chain_type="stuff")


def display_chat_history(chat_history):
    for chat in chat_history:
        background_color = "#ffeecf" if chat[2] == "new" else "#ffeecf" if chat[0] == "User" else "#ffeecf"
        st.markdown(f"<div style='background-color: {background_color}; padding: 10px; border-radius: 10px; margin: 10px;'>{chat[0]}: {chat[1]}</div>", unsafe_allow_html=True)


def handle_no_answer(response):
    no_answer_phrases = [
        "ich weiß es nicht",
        "ich weiß nicht",
        "ich bin mir nicht sicher",
        "es wird nicht erwähnt",
        "Leider kann ich diese Frage nicht beantworten",
        "kann ich diese Frage nicht beantworten",
        "ich kann diese Frage nicht beantworten",
        "ich kann diese Frage leider nicht beantworten",
        "keine information",
        "das ist unklar",
        "da habe ich keine antwort",
        "das kann ich nicht beantworten",
        "i don't know",
        "i am not sure",
        "it is not mentioned",
        "no information",
        "that is unclear",
        "i have no answer",
        "i cannot answer that",
        "unable to provide an answer",
        "not enough context",
        "Sorry, I do not have enough information",
        "I do not have enough information",
        "I don't have enough information",
        "Sorry, I don't have enough context to answer that question.",
        "I don't have enough context to answer that question.",
        "to answer that question.",
        "Sorry",
        "I'm sorry",
        "I don't understand the question",
        "I don't understand"

    ]

    alternative_responses = [
        "Hmm, das ist eine knifflige Frage. Lass uns das gemeinsam erkunden. Kannst du mehr Details geben?",
        "Interessante Frage! Ich bin mir nicht sicher, aber wir können es herausfinden. Hast du weitere Informationen?",
        "Das ist eine gute Frage. Ich habe momentan keine Antwort darauf, aber vielleicht kannst du sie anders formulieren?",
        "Da bin ich überfragt. Kannst du die Frage anders stellen oder mir mehr Kontext geben?",
        "Ich stehe hier etwas auf dem Schlauch. Gibt es noch andere Aspekte der Frage, die wir betrachten könnten?",
        # Add more alternative responses as needed
    ]

    # Check if response matches any phrase in no_answer_phrases
    if any(phrase in response.lower() for phrase in no_answer_phrases):
        return random.choice(alternative_responses)  # Randomly select a response
    return response




def page1():
    try:
        hide_streamlit_style = """
                <style>
                #MainMenu {visibility: hidden;}
                footer {visibility: hidden;}
                </style>
                """
        st.markdown(hide_streamlit_style, unsafe_allow_html=True)
    
         # Create columns for layout
        col1, col2 = st.columns([3, 1])  # Adjust the ratio to your liking

        with col1:
            st.title("KH_reform!")

        with col2:
            # Load and display the image in the right column, which will be the top-right corner of the page
            image = Image.open('BinDoc Logo (Quadratisch).png')
            st.image(image, use_column_width='always')

 
        # Start tracking user interactions
        with streamlit_analytics.track():
            if not os.path.exists(pdf_path):
                st.error("File not found. Please check the file path.")
                return
    
            VectorStore = load_vector_store(pdf_path, "KH_Reform_2301", force_reload=False)

            display_chat_history(st.session_state['chat_history_page1'])
    
            st.write("<!-- Start Spacer -->", unsafe_allow_html=True)
            st.write("<div style='flex: 1;'></div>", unsafe_allow_html=True)
            st.write("<!-- End Spacer -->", unsafe_allow_html=True)
    
            new_messages_placeholder = st.empty()
    
            query = st.text_input("Geben Sie hier Ihre Frage ein / Enter your question here:")

            add_vertical_space(2)  # Adjust as per the desired spacing
            
            # Create two columns for the buttons
            col1, col2 = st.columns(2)
            
            with col1:
                if st.button("Wie viele Ärzte benötigt eine Klinik in der Leistungsgruppe Pädiatrie?"):
                    query = "Wie viele Ärzte benötigt eine Klinik in der Leistungsgruppe Pädiatrie?"
                if st.button("Wie viele Leistungsgruppen gibt es?"):
                    query = ("Wie viele Leistungsgruppen gibt es?")
                if st.button("Was sind die hauptsächlichen Änderungsvorhaben der Krankenhausreform?"):
                    query = "Was sind die hauptsächlichen Änderungsvorhaben der Krankenhausreform?"

            
            with col2:
                if st.button("Welche und wieviele Fachärzte benötige ich für die Leistungsgruppe Gastroenterologie? "):
                    query = "Welche und wieviele Fachärzte benötige ich für die Leistungsgruppe Gastroenterologie"
                if st.button("Was soll die Reform der Notfallversorgung beinhalten?"):
                    query = "Was soll die Reform der Notfallversorgung beinhalten?"
                if st.button("Was bedeutet die Vorhaltefinanzierung?"):
                    query = "Was bedeutet die Vorhaltefinanzierung?"


        
            if query:
                st.session_state['chat_history_page1'].append(("User", query, "new"))

                # Start timing
                start_time = time.time()
                
                with st.spinner('Bot is thinking...'):
                    chain = load_chatbot()
                    docs = VectorStore.similarity_search(query=query, k=5)
                    with get_openai_callback() as cb:
                        response = chain.run(input_documents=docs, question=query)
                        response = handle_no_answer(response)  # Process the response through the new function


                        
                # Stop timing
                end_time = time.time()
                
                # Calculate duration
                duration = end_time - start_time

                # You can use Streamlit's text function to display the timing
                st.text(f"Response time: {duration:.2f} seconds")
    
                st.session_state['chat_history_page1'].append(("Bot", response, "new"))
    
    
                # Display new messages at the bottom
                new_messages = st.session_state['chat_history_page1'][-2:]
                for chat in new_messages:
                    background_color = "#ffeecf" if chat[2] == "new" else "#ffeecf" if chat[0] == "User" else "#ffeecf"
                    new_messages_placeholder.markdown(f"<div style='background-color: {background_color}; padding: 10px; border-radius: 10px; margin: 10px;'>{chat[0]}: {chat[1]}</div>", unsafe_allow_html=True)
    
    
                # Clear the input field after the query is made
                query = ""
    
            # Mark all messages as old after displaying
            st.session_state['chat_history_page1'] = [(sender, msg, "old") for sender, msg, _ in st.session_state['chat_history_page1']]

    except Exception as e:
        st.error(f"Upsi, an unexpected error occurred: {e}")
        # Optionally log the exception details to a file or error tracking service




def page2():
    try:
        hide_streamlit_style = """
                <style>
                #MainMenu {visibility: hidden;}
                footer {visibility: hidden;}
                </style>
                """
        st.markdown(hide_streamlit_style, unsafe_allow_html=True)
    
         # Create columns for layout
        col1, col2 = st.columns([3, 1])  # Adjust the ratio to your liking

        with col1:
            st.title("Kennzahlenbuch 100 Kennzahlen!")

        with col2:
            # Load and display the image in the right column, which will be the top-right corner of the page
            image = Image.open('BinDoc Logo (Quadratisch).png')
            st.image(image, use_column_width='always')

 
        # Start tracking user interactions
        with streamlit_analytics.track():
            
            if not os.path.exists(pdf_path2):
                st.error("File not found. Please check the file path.")
                return
    
            VectorStore = load_vector_store(pdf_path2, "Buch_2301", force_reload=False)
    

    
            display_chat_history(st.session_state['chat_history_page2'])
    
            st.write("<!-- Start Spacer -->", unsafe_allow_html=True)
            st.write("<div style='flex: 1;'></div>", unsafe_allow_html=True)
            st.write("<!-- End Spacer -->", unsafe_allow_html=True)
    
            new_messages_placeholder = st.empty()
    
            query = st.text_input("Ask questions about your PDF file (in any preferred language):")

            add_vertical_space(2)  # Adjust as per the desired spacing
            
            # Create two columns for the buttons
            col1, col2 = st.columns(2)
            
            with col1:
                if st.button("Nenne mir 5 wichtige Personalkennzahlen im Krankenhaus."):
                    query = "Nenne mir 5 wichtige Personalkennzahlen im Krankenhaus."
                if st.button("Wie ist die durchschnittliche Bettenauslastung eines Krankenhauses?"):
                    query = ("Wie ist die durchschnittliche Bettenauslastung eines Krankenhauses?")
                if st.button("Welches sind die häufigsten DRGs, die von den Krankenhäusern abgerechnet werden?"):
                    query = "Welches sind die häufigsten DRGs, die von den Krankenhäusern abgerechnet werden? "

            
            with col2:
                if st.button("Wieviel Casemixpunkte werden im Median von einer ärztlichen Vollkraft erbracht?"):
                    query = "Wieviel Casemixpunkte werden im Median von einer ärztlichen Vollkraft erbracht?"
                if st.button("Bitte erstelle mir einer Übersicht der wichtiger Strukturkennzahlen eines Krankenhauses der Grund- und Regelversorgung."):
                    query = "Bitte erstelle mir einer Übersicht der wichtiger Strukturkennzahlen eines Krankenhauses der Grund- und Regelversorgung."
                if st.button("Wieviele Patienten eines Grund- und Regelversorgers kommen aus welcher Fahrzeitzone?"):
                    query = "Wieviele Patienten eines Grund- und Regelversorgers kommen aus welcher Fahrzeitzone?"

        
            if query:
                st.session_state['chat_history_page2'].append(("User", query, "new"))

                # Start timing
                start_time = time.time()
                
                with st.spinner('Bot is thinking...'):
                    chain = load_chatbot()
                    docs = VectorStore.similarity_search(query=query, k=5)
                    with get_openai_callback() as cb:
                        response = chain.run(input_documents=docs, question=query)
                        response = handle_no_answer(response)  # Process the response through the new function


                        
                # Stop timing
                end_time = time.time()
                
                # Calculate duration
                duration = end_time - start_time

                # You can use Streamlit's text function to display the timing
                st.text(f"Response time: {duration:.2f} seconds")
    
                st.session_state['chat_history_page2'].append(("Bot", response, "new"))
    
    
                # Display new messages at the bottom
                new_messages = st.session_state['chat_history_page2'][-2:]
                for chat in new_messages:
                    background_color = "#ffeecf" if chat[2] == "new" else "#ffeecf" if chat[0] == "User" else "#ffeecf"
                    new_messages_placeholder.markdown(f"<div style='background-color: {background_color}; padding: 10px; border-radius: 10px; margin: 10px;'>{chat[0]}: {chat[1]}</div>", unsafe_allow_html=True)
    
    
                # Clear the input field after the query is made
                query = ""
    
            # Mark all messages as old after displaying
            st.session_state['chat_history_page2'] = [(sender, msg, "old") for sender, msg, _ in st.session_state['chat_history_page2']]

    except Exception as e:
        st.error(f"Upsi, an unexpected error occurred: {e}")
        # Optionally log the exception details to a file or error tracking service



def page3():
    try:
        hide_streamlit_style = """
                <style>
                #MainMenu {visibility: hidden;}
                footer {visibility: hidden;}
                </style>
                """
        st.markdown(hide_streamlit_style, unsafe_allow_html=True)
    
         # Create columns for layout
        col1, col2 = st.columns([3, 1])  # Adjust the ratio to your liking

        with col1:
            st.title("Kosten- und Strukturdaten der Krankenhäuser")

        with col2:
            # Load and display the image in the right column, which will be the top-right corner of the page
            image = Image.open('BinDoc Logo (Quadratisch).png')
            st.image(image, use_column_width='always')

 
        # Start tracking user interactions
        with streamlit_analytics.track():
            
            if not os.path.exists(pdf_path2):
                st.error("File not found. Please check the file path.")
                return
    
            VectorStore = load_vector_store(pdf_path3, "Kosten_Str_2301", force_reload=False)
    

    
            display_chat_history(st.session_state['chat_history_page3'])
    
            st.write("<!-- Start Spacer -->", unsafe_allow_html=True)
            st.write("<div style='flex: 1;'></div>", unsafe_allow_html=True)
            st.write("<!-- End Spacer -->", unsafe_allow_html=True)
    
            new_messages_placeholder = st.empty()
    
            query = st.text_input("Ask questions about your PDF file (in any preferred language):")

            add_vertical_space(2)  # Adjust as per the desired spacing
            
            # Create two columns for the buttons
            col1, col2 = st.columns(2)
            
            with col1:
                if st.button("Wie hat sich die Bettenanzahl in den letzten 10 Jahren entwickelt?"):
                    query = "Wie hat sich die Bettenanzahl in den letzten 10 Jahren entwickelt?"
                if st.button("Wieviele Patienten werden pro Jahr stationär in Deutschland behandelt?"):
                    query = ("Wieviele Patienten werden pro Jahr stationär in Deutschland behandelt?")
                if st.button("Wieviele Vollkräfte arbeiten in Summe in deutschen Krankenhäusern?"):
                    query = "Wieviele Vollkräfte arbeiten in Summe in deutschen Krankenhäusern? "

            
            with col2:
                if st.button("Welche unterschiedlichen Personalkosten gibt es im Krankenhaus?"):
                    query = "Welche unterschiedlichen Personalkosten gibt es im Krankenhaus?"
                if st.button("Welche Sachkosten werden in Krankenhäusern unterschieden?"):
                    query = "Welche Sachkosten werden in Krankenhäusern unterschieden? "
                if st.button("Wie hoch sind die Gesamtkosten der Krankenhäuser pro Jahr?"):
                    query = "Wie hoch sind die Gesamtkosten der Krankenhäuser pro Jahr?"

        
            if query:
                st.session_state['chat_history_page3'].append(("User", query, "new"))

                # Start timing
                start_time = time.time()
                
                with st.spinner('Bot is thinking...'):
                    chain = load_chatbot()
                    docs = VectorStore.similarity_search(query=query, k=5)
                    with get_openai_callback() as cb:
                        response = chain.run(input_documents=docs, question=query)
                        response = handle_no_answer(response)  # Process the response through the new function


                        
                # Stop timing
                end_time = time.time()
                
                # Calculate duration
                duration = end_time - start_time

                # You can use Streamlit's text function to display the timing
                st.text(f"Response time: {duration:.2f} seconds")
    
                st.session_state['chat_history_page3'].append(("Bot", response, "new"))
    
    
                # Display new messages at the bottom
                new_messages = st.session_state['chat_history_page3'][-2:]
                for chat in new_messages:
                    background_color = "#ffeecf" if chat[2] == "new" else "#ffeecf" if chat[0] == "User" else "#ffeecf"
                    new_messages_placeholder.markdown(f"<div style='background-color: {background_color}; padding: 10px; border-radius: 10px; margin: 10px;'>{chat[0]}: {chat[1]}</div>", unsafe_allow_html=True)
    
    
                # Clear the input field after the query is made
                query = ""
    
            # Mark all messages as old after displaying
            st.session_state['chat_history_page3'] = [(sender, msg, "old") for sender, msg, _ in st.session_state['chat_history_page3']]

    except Exception as e:
        st.error(f"Upsi, an unexpected error occurred: {e}")
        # Optionally log the exception details to a file or error tracking service



def main():
    # Sidebar content
    with st.sidebar:
        st.title('BinDoc GmbH')
        st.markdown("Experience revolutionary interaction with BinDocs Chat App, leveraging state-of-the-art AI technology.")
        add_vertical_space(1)
        page = st.sidebar.selectbox("Choose a page", ["KH_Reform", "Kennzahlenbuch 100 Kennzahlen", "Kosten- und Strukturdaten der Krankenhäuser"])
        add_vertical_space(1)
        st.write('Made with ❤️ by BinDoc GmbH')

    # Main area content based on page selection
    if page == "KH_Reform":
        page1()
    elif page == "Kennzahlenbuch 100 Kennzahlen":
        page2()
    elif page == "Kosten- und Strukturdaten der Krankenhäuser":
        page3()


if __name__ == "__main__":
    main()