app.py

import os
from datasets import load_dataset
import random
import pickle
from nltk.tokenize import sent_tokenize
import nltk
from PyPDF2 import PdfReader
import streamlit as st
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

nltk.download('punkt')



# Sidebar contents
with st.sidebar:
    st.title(':orange_book: BinDoc GmbH')
    

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

    if not api_key:
        st.warning('API key is required to proceed.')
        st.stop()  # Stop the app if the API key is not provided

    st.markdown("Experience the future of document interaction with the revolutionary")
    st.markdown("**BinDocs Chat App**.")
    st.markdown("Harnessing the power of a Large Language Model and AI technology,")
    st.markdown("this innovative platform redefines PDF engagement,")
    st.markdown("enabling dynamic conversations that bridge the gap between")
    st.markdown("human and machine intelligence.")

    add_vertical_space(3)  # Add more vertical space between text blocks
    st.write('Made with ❤️ by BinDoc GmbH')

def load_pdf(file_path):
    pdf_reader = PdfReader(file_path)
    chunks = []
    for page in pdf_reader.pages:
        text = page.extract_text()
        if text:
            chunks.append(text)
    
    store_name = file_path.name[:-4]
    
    if os.path.exists(f"{store_name}.pkl"):
        with open(f"{store_name}.pkl", "rb") as f:
            VectorStore = pickle.load(f)
    else:
        embeddings = OpenAIEmbeddings()
        VectorStore = FAISS.from_texts(chunks, embedding=embeddings)
        with open(f"{store_name}.pkl", "wb") as f:
            pickle.dump(VectorStore, f)

    return VectorStore

def load_chatbot(max_tokens=300):
    return load_qa_chain(llm=OpenAI(temperature=0.1, max_tokens=max_tokens), chain_type="stuff") 
    

def display_chat_history(chat_history):
    for chat in chat_history:
        background_color = "#FFA07A" if chat[2] == "new" else "#acf" if chat[0] == "User" else "#caf"
        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 remove_incomplete_sentences(text):
    sentences = sent_tokenize(text)
    complete_sentences = [sent for sent in sentences if sent.endswith(('.', '!', '?'))]
    return ' '.join(complete_sentences)

def remove_redundant_information(text):
    sentences = sent_tokenize(text)
    unique_sentences = list(set(sentences))
    return ' '.join(unique_sentences)

# Define a maximum token limit to avoid infinite loops
MAX_TOKEN_LIMIT = 400

import random


def main():
    st.title("BinDocs Chat App")

    # Step 1: Adding CSS for rounded boxes
    st.markdown("""
        <style>
            .question-box {
                border: 1px solid orange;
                border-radius: 15px;
                padding: 10px;
                text-align: center;
                cursor: pointer;
                display: inline-block;
                width: 45%;
                margin: 2%;
            }
        </style>
    """, unsafe_allow_html=True)

    if "chat_history" not in st.session_state:
        st.session_state['chat_history'] = []

    display_chat_history(st.session_state['chat_history'])

    new_messages_placeholder = st.empty()

    pdf = st.file_uploader("Upload your PDF", type="pdf")

    query = st.text_input("Ask questions about your PDF file (in any preferred language):")

    if st.button("Ask") or (query and query != st.session_state.get('last_input', '')):
        if pdf is not None:
            st.session_state['last_input'] = query
            st.session_state['chat_history'].append(("User", query, "new"))

            loading_message = st.empty()
            loading_message.text('Bot is thinking...')
            
            VectorStore = load_pdf(pdf)
            max_tokens = 120
            chain = load_chatbot(max_tokens=max_tokens)
            docs = VectorStore.similarity_search(query=query, k=2)
            
            with get_openai_callback() as cb:
                response = chain.run(input_documents=docs, question=query)

            # Post-processing to remove incomplete sentences and redundant information
            filtered_response = remove_incomplete_sentences(response)
            filtered_response = remove_redundant_information(filtered_response)
            
            st.session_state['chat_history'].append(("Bot", filtered_response, "new"))

            new_messages = st.session_state['chat_history'][-2:]
            for chat in new_messages:
                background_color = "#FFA07A" if chat[2] == "new" else "#acf" if chat[0] == "User" else "#caf"
                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)

            st.write("<script>document.getElementById('response').scrollIntoView();</script>", unsafe_allow_html=True)

            loading_message.empty()

            query = ""
        else:
            st.warning("Please upload a PDF file before asking questions.")

        st.session_state['chat_history'] = [(sender, msg, "old") for sender, msg, _ in st.session_state['chat_history']]
    


    
    # Displaying example questions
    if not st.session_state['chat_history']:
        st.markdown("""
            <div class="question-box" id="question1">Was genau ist ein Belegarzt?</div>
            <div class="question-box" id="question2">Wofür wird die Alpha-ID verwendet?</div>
            <br>
            <div class="question-box" id="question3">Was sind die Vorteile des ambulanten operierens?</div>
        """, unsafe_allow_html=True)

     st.markdown("""
        <script>
            document.getElementById('question1').onclick = function() {
                console.log('Question 1 box clicked');
            };
            document.getElementById('question2').onclick = function() {
                console.log('Question 2 box clicked');
            };
            document.getElementById('question3').onclick = function() {
                console.log('Question 3 box clicked');
            };
        </script>
    """, unsafe_allow_html=True)




if __name__ == "__main__":
    main()