utils/chatbot.py

import os
import pandas as pd
import numpy as np
from utils.openai_api import get_response
from utils.document_retrieval import select_document_section_by_query_similarity
from utils.file_utils import load_file

# Load the prompts from the files
meta_prompt_file_path = '/home/user/app/prompts/meta_prompt.txt'
end_message_file_path = '/home/user/app/prompts/end_message.txt'
META_PROMPT = load_file(meta_prompt_file_path)
end_message = load_file(end_message_file_path)

def construct_prompt(query: str, context_embeddings: dict, df: pd.DataFrame) -> tuple[str, str]:
    """
    Constructs a prompt for the language model based on the most relevant service description.

    This function identifies the most relevant service by comparing the query with precomputed 
    document embeddings. It then formats the prompt to include an introduction, the service 
    description as context, and the user's question.

    Parameters:
        query (str): The user's input question.
        context_embeddings (dict): A dictionary mapping service identifiers to their embeddings.
        df (pd.DataFrame): A DataFrame containing service descriptions and links.

    Returns:
        tuple[str, str]: A tuple containing the formatted prompt and the associated service link.
    """
    # Select the most relevant service based on the query
    _, chosen_service = select_document_section_by_query_similarity(query, context_embeddings)
        
    # Format the service description and clean up newline characters
    service_description = df.loc[chosen_service].description.replace("\n", " ")
    
    # Construct the introduction and the full prompt
    introduction = (
        "Answer the question as truthfully as possible using the provided context. "
        "If the answer is not contained within the text below, say: "
        "'I could not find an answer to your question, please reach out to Helpdesk.'"
    )
    
    question = f"\n\nQ: {query}"
    message = f"{introduction}\n* \n\nContext:\n{service_description}{question}"
    
    # Get the relevant service link
    link = df.loc[chosen_service].link
    
    return message, link

def answer_query_with_context(
    query: str,
    df: pd.DataFrame,
    document_embeddings: dict[(str, str), np.array],
    META_PROMPT=META_PROMPT,
    end_message=end_message
) -> str:
    """
    Generates a response to a user's query using the most relevant service description.

    This function constructs a prompt by attaching the most relevant service description to the user's query,
    sends this prompt to the language model to generate a response, and appends additional service-related 
    information to this response.

    Parameters:
        query (str): The user's input question.
        df (pd.DataFrame): A DataFrame containing service descriptions and links.
        document_embeddings (dict): A dictionary mapping service identifiers to their embeddings.
        show_prompt (bool, optional): If True, displays the constructed prompt (for debugging). Defaults to False.

    Returns:
        str: The final response from the chatbot, including the generated answer and additional service details.
    """
    # Construct the prompt and retrieve the service link
    prompt, link = construct_prompt(query, document_embeddings, df)
    
    # Get the response from the language model
    messages=[
        {"role": "system", "content": META_PROMPT},
        {"role": "user", "content": prompt}
        ]
    
    response = get_response(messages)  
    print(end_message)

    return response + end_message