app.py

import os
from dotenv import load_dotenv
from PyPDF2 import PdfReader
from langchain.text_splitter import CharacterTextSplitter
from langchain import vectorstores
from langchain import chains
from langchain import llms
from langchain.embeddings import HuggingFaceEmbeddings
import gradio as gr

llm = llms.AI21(ai21_api_key='diNNQzvL40ZnBnEQkIBwNESWjtj792NG')

def pdf_qa(pdf, query):
    if pdf is not None:
        pdf_reader = PdfReader(pdf)
        texts = ""
        for page in pdf_reader.pages:
            texts += page.extract_text()
        text_splitter = CharacterTextSplitter(
            separator="\n",
            chunk_size=1000,
            chunk_overlap=0
        )
        chunks = text_splitter.split_text(texts)
        embeddings = HuggingFaceEmbeddings()
        db = vectorstores.Chroma.from_texts(chunks, embeddings)
        retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 10})
        qa = chains.ConversationalRetrievalChain.from_llm(llm=llm, retriever=retriever)
        chat_history = []
        if query:
            result = qa({"question": query, "chat_history": chat_history})
            return result["answer"]
    return "Please upload a PDF and enter a query."

with gr.Blocks() as demo:
    pdf_input = gr.File(label="Upload your PDF", type="file")
    query_input = gr.Textbox(label="Ask a question in PDF")
    output = gr.Textbox(label="Answer")
    submit_button = gr.Button("Submit")

    submit_button.click(fn=pdf_qa, inputs=[pdf_input, query_input], outputs=output)

demo.launch()