import os
import pandas as pd
import requests
from flask import Flask, request, jsonify
from flask_cors import CORS
from fuzzywuzzy import fuzz, process

app = Flask(__name__)
CORS(app)

# --- Configuration ---
# Get the Hugging Face API token from the Space's secrets
HF_API_TOKEN = os.environ.get("HF_API_TOKEN")
# The URL for the Sentence Similarity Inference API
API_URL = "https://api-inference.huggingface.co/models/sentence-transformers/paraphrase-MiniLM-L6-v2"
KNOWLEDGE_BASE_PATH = 'dataset.csv'

# --- Load Data Once on Startup ---
# This is lightweight and will not cause a crash
try:
    df = pd.read_csv(KNOWLEDGE_BASE_PATH)
    df.dropna(subset=['questions', 'answers'], inplace=True)
    QUESTIONS_LIST = df['questions'].tolist()
    print("--- Knowledge base loaded successfully. ---")
except FileNotFoundError:
    df = None
    QUESTIONS_LIST = []
    print(f"FATAL: Knowledge base file not found at {KNOWLEDGE_BASE_PATH}")

def get_greeting_response(text):
    greetings = {
        "hello": "Hello! 👋 How can I help you today?", "hi": "Hi there! How can I assist you?",
        "hey": "Hey! What can I do for you?", "how are you": "I'm just a bot, but I'm ready to help! 😊",
        "thank you": "You're welcome! Is there anything else I can help with?", "thanks": "You're welcome!",
        "bye": "Thank you for chatting. Goodbye! 👋", "goodbye": "Goodbye! Have a great day. 👋"
    }
    text_lower = text.lower()
    for greet, response in greetings.items():
        if fuzz.partial_ratio(greet, text_lower) > 85:
            return response
    return None

def get_bot_response(user_query):
    # First, check for simple greetings
    greeting = get_greeting_response(user_query)
    if greeting:
        return greeting

    # Then, check for a very close fuzzy match to avoid unnecessary API calls
    best_match, score = process.extractOne(user_query, QUESTIONS_LIST, scorer=fuzz.token_sort_ratio)
    if score > 85:
        idx = QUESTIONS_LIST.index(best_match)
        return df.iloc[idx]['answers']

    # If no close match, use the powerful Sentence Similarity API
    if not HF_API_TOKEN:
        return "Error: Hugging Face API token is not configured."

    headers = {"Authorization": f"Bearer {HF_API_TOKEN}"}
    payload = {
        "inputs": {
            "source_sentence": user_query,
            "sentences": QUESTIONS_LIST
        }
    }

    try:
        response = requests.post(API_URL, headers=headers, json=payload, timeout=20)
        response.raise_for_status() # Raise an exception for bad status codes (4xx or 5xx)
        scores = response.json()

        if scores and isinstance(scores, list):
            # Find the index of the highest score
            best_score_index = scores.index(max(scores))
            # If the best score is above a certain threshold, return the answer
            if scores[best_score_index] > 0.5:
                return df.iloc[best_score_index]['answers']

    except requests.exceptions.RequestException as e:
        print(f"API Request Error: {e}")
        return "Sorry, I'm having trouble reaching my knowledge circuits right now."

    return "I'm sorry, I don't have information on that topic. Could you try asking in a different way?"

@app.route('/api/chat', methods=['POST'])
def chat():
    if df is None:
        return jsonify({'error': 'Service is unavailable due to a missing knowledge base.'}), 503

    try:
        data = request.get_json()
        user_message = data.get('message')
        if not user_message:
            return jsonify({'error': 'No message provided'}), 400

        bot_response = get_bot_response(user_message)
        return jsonify({'response': bot_response})
    except Exception as e:
        print(f"Error in /api/chat: {e}")
        return jsonify({'error': 'An internal server error occurred'}), 500