app.py

import google.generativeai as palm
import pandas as pd

import os

import gradio as gr
import io

from langchain.llms import GooglePalm
import pandas as pd
#from yolopandas import pd

from langchain.embeddings import GooglePalmEmbeddings
# a class to create a question answering system based on information retrieval
from langchain.chains import RetrievalQA
# a class for splitting text into fixed-sized chunks with an optional overlay
from langchain.text_splitter import RecursiveCharacterTextSplitter
# a class to create a vector index using FAISS, a library for approximate nearest neighbor search
from langchain.vectorstores import FAISS
# a class for loading PDF documents from a directory
from langchain.document_loaders import PyPDFDirectoryLoader
from langchain.chains.question_answering import load_qa_chain
from langchain.chains import ConversationalRetrievalChain
from langchain.schema.vectorstore import VectorStoreRetriever

from dotenv import load_dotenv

load_dotenv()




palm.configure(api_key=os.environ['PALM'])

models = [m for m in palm.list_models(
) if 'generateText' in m.supported_generation_methods]
model = models[0].name
print(model)

def get_pdf_text(pdf_docs):
    text=""
    for pdf in pdf_docs:
        pdf_reader= PdfReader(pdf)
        for page in pdf_reader.pages:
            text+= page.extract_text()
    return  text

# load PDF files from a directory
loader = PyPDFDirectoryLoader("documents/")
data = loader.load()

# print the loaded data, which is a list of tuples (file name, text extracted from the PDF)
#print(data)

# split the extracted data into text chunks using the text_splitter, which splits the text based on the specified number of characters and overlap
text_splitter = RecursiveCharacterTextSplitter(chunk_size=10000, chunk_overlap=20)

text_chunks = text_splitter.split_documents(data)

# print the number of chunks obtained
#print(len(text_chunks))

embeddings = GooglePalmEmbeddings(google_api_key=os.environ['PALM'])

# create embeddings for each text chunk using the FAISS class, which creates a vector index using FAISS and allows efficient searches between vectors
vector_store = FAISS.from_documents(text_chunks, embedding=embeddings)

#print(type(vector_store))



def ask_pdfs(user_question):
    load_dotenv()



    llm = GooglePalm(temperature=0, google_api_key=os.environ['PALM'])

    # Create a question answering system based on information retrieval using the RetrievalQA class, which takes as input a neural language model, a chain type and a retriever (an object that allows you to retrieve the most relevant chunks of text for a query)
    retriever = VectorStoreRetriever(vectorstore=vector_store)
    #qa = RetrievalQA.from_llm(llm=llm, retriever=retriever, return_source_documents=True)
    qa2 = RetrievalQA.from_llm(llm=llm, retriever=retriever)
    #response =qa(user_question)
    res = qa2.run(user_question)
    #print("Response:",response)
    #ans = str(res + "Source document:"+ str(response['source_documents']))
    return res





'''

def questiondocument(user_question):
    load_dotenv()



    llm = GooglePalm(temperature=0, google_api_key=os.environ['PALM'])
    response = llm(user_question)

    return response
'''
demo = gr.Interface(

    fn=ask_pdfs,
    inputs=["text"],
    outputs=["text"],
    title="UUW QuickHelper Bot",
)
demo.launch(share=True)