app.py

# Importing the libraries
from dotenv import load_dotenv
import streamlit as st  # for the web app
from PyPDF2 import PdfReader  # for reading the pdf
from dotenv import load_dotenv  # for loading the environment variables
# for splitting the text into sentences
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import FAISS  # for storing the vectors
from langchain.llms import OpenAI  # for the language model
# for loading the question answering chain
from langchain.chains.question_answering import load_qa_chain
from langchain.callbacks import get_openai_callback  # for getting the callback
import pickle  # for loading the pickle files
import os  # for the environment variables
from langchain.chat_models import ChatOpenAI  # for the chat model


# Building the UI components using Streamlit
# Streamlit is a Python library that makes it easy to build beautiful apps for machine learning
st.header('ChatPDF v0.1')
st.sidebar.header(":blue[Welcome to ChatPDF!]")
pdf = st.file_uploader(
    'Upload a PDF file with text in English. PDFs that only contain images will not be recognized.', type=['pdf'])
query = st.text_input('Ask question about the PDF you entered!', max_chars=300)

txt = ""  # Initialize txt as an empty string

# Using PyPDF2 to read the pdf
try:
    pdf_doc = PdfReader(pdf)
    for page in pdf_doc.pages:  # for each page in the pdf
        # extract the text from the page and add it to the txt variable
        txt += page.extract_text()

except Exception as e:
    st.error(str(e))  # if there is an error, print the error

# Using the RecursiveCharacterTextSplitter to split the text into sentences
text_split = RecursiveCharacterTextSplitter(
    chunk_size=1000,  # number of characters per chunk
    # used to keep the context of a chunk intact with previous and next chunks
    chunk_overlap=200,
    length_function=len
)
chunks = text_split.split_text(text=txt)

load_dotenv()

# Loading the embeddings and the vector store using the FAISS library
embeddings = OpenAIEmbeddings(openai_api_key=os.environ["OPENAI_API_KEY"])
vectorStore = FAISS.from_texts(chunks, embedding=embeddings)

print("Length of chunks:", len(chunks))
for i, chunk in enumerate(chunks):
    print(f"Chunk {i + 1}:", chunk)
# Add more details about the embeddings object if needed
print("Embeddings object:", embeddings)

# Search on user's input query and return the most similar sentence
# gpt-3.5-turbo is used as the language model because it is cost effective and has a good performance
docs = vectorStore.similarity_search(query=query)
llm = ChatOpenAI(model_name="gpt-3.5-turbo")

chain = load_qa_chain(llm=llm, chain_type="stuff") #loading the question answering chain
#This limits it to response only if it knows the answer and try not to make up an answer

#Running the chain
response = chain.run(input_documents=docs, question=query)
st.write(response)

# Saving the vector store
with open(f"STORE_NAME.pkl", "wb") as f:
  pickle.dump(vs, f)

# Loading the vector store
with open(f"{store_name}.pkl", "rb") as f:
  vs = pickle.load(f)