generate_answer.py

import os
from glob import glob
import openai
from dotenv import load_dotenv

from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter

from langchain_community.chat_models import ChatOpenAI
from langchain.chains import RetrievalQA
from langchain.memory import ConversationBufferMemory

load_dotenv()
api_key = os.getenv("OPENAI_API_KEY")
openai.api_key = api_key

# Helper function to validate response completeness
def is_response_complete(response: str) -> bool:
    return response.strip()[-1] in ".!?"

# Retry mechanism for incomplete responses
def retry_response(messages):
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=messages
    ).choices[0].message['content']
    if not is_response_complete(response):
        response += " This is the end of the response. Please let me know if you need further clarification."
    return response

def base_model_chatbot(messages):
    system_message = [
        {"role": "system", "content": "You are a helpful AI chatbot that provides clear, complete, and coherent responses to User's questions. Ensure your answers are in full sentences and complete the thought or idea."}
    ]
    messages = system_message + messages
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=messages
    ).choices[0].message['content']
    # Validate response completeness
    if not is_response_complete(response):
        response = retry_response(messages)
    return response

class VectorDB:
    """Class to manage document loading and vector database creation."""
    
    def __init__(self, docs_directory: str):
        self.docs_directory = docs_directory

    def create_vector_db(self):
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)

        files = glob(os.path.join(self.docs_directory, "*.pdf"))

        loadPDFs = [PyPDFLoader(pdf_file) for pdf_file in files]

        pdf_docs = list()
        for loader in loadPDFs:
            pdf_docs.extend(loader.load())
        chunks = text_splitter.split_documents(pdf_docs)
            
        return Chroma.from_documents(chunks, OpenAIEmbeddings()) 
    
class ConversationalRetrievalChain:
    """Class to manage the QA chain setup."""

    def __init__(self, model_name="gpt-3.5-turbo", temperature=0):
        self.model_name = model_name
        self.temperature = temperature

    def create_chain(self):
        model = ChatOpenAI(
            model_name=self.model_name, 
            temperature=self.temperature,
            system_prompt="You are a knowledgeable AI that answers questions based on provided documents. Always give responses in clear, complete sentences."
        )
        memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
        vector_db = VectorDB('docs/')
        retriever = vector_db.create_vector_db().as_retriever(search_type="similarity", search_kwargs={"k": 2})
        return RetrievalQA.from_chain_type(
            llm=model,
            retriever=retriever,
            memory=memory,
        )
    
def with_pdf_chatbot(messages):
    query = messages[-1]['content'].strip()
    qa_chain = ConversationalRetrievalChain().create_chain()
    result = qa_chain({"query": query})
    if not is_response_complete(result['result']):
        result['result'] += " This is the end of the response. Let me know if you need further clarification."
    return result['result']