import os,re
import gradio as gr
import nest_asyncio
from langchain import PromptTemplate
from llama_index.core import PromptTemplate, StorageContext, load_index_from_storage
from llama_index.llms.openai import OpenAI

import networkx as nx
from pyvis.network import Network
from IPython.display import HTML, Markdown, display


# Define constants for index paths
os.environ["OPENAI_API_KEY"] = os.getenv('oai')
KG_INDEX_PATH = "./telcom_full_property_kg_processed_dynamicpath2_withpackagedata_category/"
KG_PLOT_PATH = KG_INDEX_PATH + "/full_kg.html"
RAG_INDEX_PATH = "./telcom_RAG_full_withpackagedata_category/"

# Load Graph-RAG index
graph_rag_index = load_index_from_storage(
    StorageContext.from_defaults(persist_dir=KG_INDEX_PATH)
)

# Load RAG index
rag_index = load_index_from_storage(
    StorageContext.from_defaults(persist_dir=RAG_INDEX_PATH)
)

def query_tqa(query, search_level):
    """
    Query both Graph-RAG and RAG models and return their responses and references.

    Args:
        query (str): The query to be processed.
        search_level (str): The level of search to be performed.

    Returns:
        tuple: Responses and references from both Graph-RAG and RAG models.
    """
    grag_response, grag_reference, grag_reference_text = query_graph_rag_qa(query, search_level)
    rag_response, rag_reference, rag_reference_text = query_rag_qa(query, search_level)
    return grag_response, grag_reference, grag_reference_text, rag_response, rag_reference, rag_reference_text


def eval_llm(query, rag_response, grag_response):
    """
    Evaluate responses from both Graph-RAG and RAG models using an LLM.

    Args:
        query (str): The query that was used to generate the responses.
        rag_response (str): The response from the RAG model.
        grag_response (str): The response from the Graph-RAG model.

    Returns:
        tuple: Evaluation results for both responses.
    """
    grag_eval = evaluate_llm(query, grag_response)
    rag_eval = evaluate_llm(query, rag_response)
    return grag_eval, rag_eval


 
with gr.Blocks() as demo:
    gr.Markdown("<h1>Telcom Graph-RAG v0.1</h1>")

    with gr.Tab("Virtual Assistant"):
        with gr.Row():
            query_input = gr.Textbox(label="Input Your Query..")
            search_level = gr.Slider(minimum=1, maximum=50, value=3, step=5, label="Search level")
            ask_button = gr.Button("Ask TelcomVA!!")

            
        with gr.Row():
            with gr.Accordion("Graph-RAG!", open=True):
                grag_output = gr.Textbox(label="Response")
                grag_reference = gr.Textbox(label="Triplets")
                grag_reference_text = gr.Textbox(label="Extracted Reference raw")

            with gr.Accordion("RAG", open=True):
                rag_output = gr.Textbox(label="Response")
                rag_reference = gr.Textbox(label="Extracted Reference")
                rag_reference_text = gr.Textbox(label="Extracted Reference raw")


        with gr.Row():
            grag_performance = gr.Textbox(label="Graph-RAG Performance")
            rag_performance = gr.Textbox(label="RAG Performance")
        eval_button = gr.Button("Evaluate LLMs!!")



    with gr.Accordion("Explore KG!", open=False):
        kg_output = gr.HTML()
        plot_button = gr.Button("Plot Full KG!!")

    ask_button.click(query_tqa, 
                     inputs=[query_input,search_level], 
                     outputs=[
                         grag_output,
                         grag_reference,
                         grag_reference_text,
                         rag_output,
                         rag_reference,
                         rag_reference_text
                         ]
    )
    
    eval_button.click(eval_llm,
                      inputs=[query_input,rag_output,grag_output],
                      outputs=[
                          grag_performance,
                          rag_performance
                      ]
    )

    plot_button.click(plot_full_kg, outputs=kg_output)

    examples = gr.Examples(
    examples=[
        ["what are the upselling ideas for roaming package you can recommend for customer Rina Wati."],
    ],
    inputs=[query_input]
    )
demo.launch(auth=(os.getenv('id'), os.getenv('pass')), share=True)