backend.py

import os
from langchain.document_loaders import UnstructuredPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings
from langchain.llms import OpenAI
from langchain.chains.question_answering import load_qa_chain
from PyPDF2 import PdfReader

def comp_process(apikey, pdfs, question):

    # Initialize our Language Model
    os.environ["OPENAI_API_KEY"] = apikey
    llm = OpenAI(temperature=0, openai_api_key=apikey)
    
    text = ""

    for file in pdfs:
        pdf_reader = PdfReader(file)
        for page in pdf_reader.pages:
            text += page.extract_text()

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    chunks = text_splitter.split_text(text=text)

    embeddings = OpenAIEmbeddings(openai_api_key=apikey)
    docsearch = Chroma.from_texts(chunks, embedding=embeddings).as_retriever()

    if question:
        docs = docsearch.get_relevant_documents(question)
        read_chain = load_qa_chain(llm=llm)
        answer = read_chain.run(input_documents=docs, question=question)

    return(answer)