app.py

from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores.faiss import FAISS
from langchain.chains import VectorDBQAWithSourcesChain
from langchain import OpenAI

import openai
import os

import gradio as gr
from pathlib import Path
import pdfplumber

os.environ["OPENAI_API_KEY"] = os.environ.get("openaiapi")


def get_info(pdf_obj, query):

    if isinstance(pdf_obj, list):
        pdf_obj = pdf_obj[0]
  
    file_path = Path(pdf_obj.name)

    with pdfplumber.open(file_path) as pdf:
        output = ''
        for page in pdf.pages:
            extract = page.extract_text()
            if len(extract) > 2048:
                output += extract[0:len(extract)//2]
                output += '\n\nNEW PAGE\n\n'
                output += extract[len(extract)//2:len(extract)]
                output += '\n\nNEW PAGE\n\n'
            else:
                output += extract
                output += '\n\nNEW PAGE\n\n'
    
    text_splitter = CharacterTextSplitter(chunk_size=2048, chunk_overlap=0, separator='\n\nNEW PAGE\n\n')
    texts = text_splitter.split_text(output.strip())

    embeddings = OpenAIEmbeddings()
    docsearch = FAISS.from_texts(texts, embeddings)
    
    # Add in a fake source information
    for i, d in enumerate(docsearch.docstore._dict.values()):
        d.metadata = {'source': f"{i}-pl"}

    chain = VectorDBQAWithSourcesChain.from_chain_type(OpenAI(temperature=0), chain_type="stuff", vectorstore=docsearch)

    return chain({"question": query}, return_only_outputs=True)['answer']


demo = gr.Blocks()

with demo:
    gr.Markdown("# Legal AI Q/A Bot😄")

    with gr.Row():

        with gr.Column():
          uploaded_file = gr.File(label="Upload a PDF file", file_count="single", type="file")
          #uploaded_file = gr.Files(label="Upload a PDF file(s)", type="file")
          input_1 = gr.Textbox(lines=1, label="Search or Ask a ques?")
          convert_button = gr.Button(variant="primary")

        with gr.Column():
            output_1 = gr.components.Textbox(label="Output")


    convert_button.click(fn=get_info, inputs=[uploaded_file, input_1], outputs=[output_1])

# demo.launch(enable_queue=True)
demo.launch()