import streamlit as st
from dotenv import load_dotenv
import pickle
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

with st.sidebar:
    st.title('PDF Chat App')
    st.markdown('''
    ## About
    This app is an LLM-powered PDF chatbot built using:
    - [Streamlit](https://streamlit.io/)
    - [LangChain](https://python.langchain.com/)
    - [OpenAI](https://platform.openai.com/docs/models) LLM model

    ## How it works
    - Load up a PDF file
    - Extract the text from the PDF file
    - Split the text into chunks
    - Create embeddings using OpenAI, which are vectors of floating-point numbers that measure the relatedness of text strings
    - Save these embeddings as vectors in a vector store, such as FAISS
    - Use a similarity search to ask a question
    - Get the answer and tokens used from OpenAI
 
    ''')
    st.write('Made with 🤖 by [Cazimir Roman](https://cazimir.dev)')

def load_app():
    # upload a PDF file
    pdf = st.file_uploader("Upload your PDF", type='pdf')

    if pdf is not None:
        pdf_reader = PdfReader(pdf)
        
        text = ""
        for page in pdf_reader.pages:
            text += page.extract_text()
        
        text_splitter = RecursiveCharacterTextSplitter(
            chunk_size = 1000,
            chunk_overlap=200,
            length_function=len
        )

        chunks = text_splitter.split_text(text=text)

        store_name = pdf.name[:-4]
        
        # check if vector store exists. if not, create one
        if os.path.exists(f"{store_name}.pkl"):
            with open(f"{store_name}.pkl", "rb") as f:
                vectorStore = pickle.load(f)
            st.success('Text embeddings loaded from disk')
        else:
            with st.spinner("Creating vector store embeddings..."):
                embeddings = OpenAIEmbeddings()
                vectorStore = FAISS.from_texts(chunks, embeddings)
                with open(f"{store_name}.pkl", "wb") as f:
                    pickle.dump(vectorStore, f)
                st.success('Embeddings computation completed')
        
        # Accept user question/query
        st.divider()
        query = st.text_input("Ask a question about your PDF file")

        if query:
            st.write(f"You asked: {query}")
            with st.spinner("Thinking..."):
                # top 3 that are most similar to our query
                docs = vectorStore.similarity_search(query)
                llm = OpenAI(temperature=0)
                chain = load_qa_chain(llm=llm, chain_type="stuff")
                with get_openai_callback() as cb:
                    response = chain.run(input_documents=docs, question=query)
                    st.write(response)

def main():
    print("Main called")
    st.header("Chat with your PDF")

    container = st.container()

    with container:
        open_ai_key = os.getenv("OPENAI_API_KEY")  
        api_key = container.text_input("Enter your OpenAI API key", type="password", value="" if open_ai_key == None else open_ai_key)
        # You can find it here: https://platform.openai.com/account/api-keys
        submit = container.button("Submit")
        
        if open_ai_key:
            load_app()

        # submit button is pressed
        if submit:
            # check if api key length correct
                if len(api_key) == 51:
                    os.environ["OPENAI_API_KEY"] = api_key
                    load_app()
                else:
                    st.error("Api key is not correct")

if __name__ == '__main__':
    main()