Agent.py

import os
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain_community.document_loaders import PyMuPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import faiss
from langchain.chains import ConversationalRetrievalChain
from langchain_community.llms import openai

class DocumentAgent:
    def __init__(self, openai_api_key: str) -> None:
        """
        Initialize the document interaction agent with necessary configurations.
        
        Parameters:
        - openai_api_key (str): API key for OpenAI services.
        """
        self.embedder = OpenAIEmbeddings(openai_api_key=openai_api_key)
        self.document_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
        self.language_model = openai.OpenAI(temperature=0, openai_api_key=openai_api_key)
        self.conversation_history = []
        self.interaction_chain = None
        self.vector_storage = None

    def respond_to_query(self, query: str) -> str:
        """
        Respond to a user query based on loaded documents.
        
        Parameters:
        - query (str): User's query.
        
        Returns:
        - str: Response generated by the agent.
        """
        if self.interaction_chain is None:
            return "Please upload a document first."
        else:
            response = self.interaction_chain({"question": query, "chat_history": self.conversation_history})
            response_text = response["answer"].strip()
            self.conversation_history.append((query, response_text))
            return response_text

    def load_document(self, document_path: str) -> None:
        """
        Load and process a document for interaction.
        
        Parameters:
        - document_path (str): Path to the document file.
        """
        loader = PyMuPDFLoader(document_path)
        documents = loader.load()
        processed_documents = self.document_splitter.split_documents(documents)

        if self.vector_storage is None:
            self.vector_storage = faiss.FAISS.from_documents(processed_documents, self.embedder)
            self.interaction_chain = ConversationalRetrievalChain.from_llm(self.language_model, self.vector_storage.as_retriever())
            self.conversation_history = []
        else:
            self.vector_storage.add_documents(processed_documents)

    def reset_agent(self) -> None:
        """
        Reset the agent's state, clearing all loaded documents and history.
        """
        self.vector_storage = None
        self.interaction_chain = None
        self.conversation_history = []