from smolagents import CodeAgent, DuckDuckGoSearchTool, tool, load_tool, HfApiModel
import requests
from bs4 import BeautifulSoup
import pandas as pd
import yaml
import datetime
import pytz
from textblob import TextBlob  # Simple sentiment analysis
from Gradio_UI import GradioUI  # For building Gradio UI

#### TOOL 1: Scrape GSMArena Reviews & Specs ####
@tool
def scrape_gsmarena_reviews(phone_name: str) -> dict:
    """Scrapes GSMArena for reviews and specs of a given smartphone.
    Args:
        phone_name: The model name (exact/substantial match).
    Returns:
        Dictionary with keys: 'reviews', 'price', 'specs'
    """
    # You’ll need accurate GSMArena URL fetching logic per phone_name.
    search_url = f"https://www.gsmarena.com/res.php3?sSearch={phone_name.replace(' ', '+')}"
    res = requests.get(search_url)
    soup = BeautifulSoup(res.text, "html.parser")
    # Find first matching phone page link
    link = soup.find("div", class_="makers").find("a")["href"]
    phone_url = f"https://www.gsmarena.com/{link}"
    phone_page = requests.get(phone_url)
    phone_soup = BeautifulSoup(phone_page.text, "html.parser")
    # Parse reviews and price
    reviews = []
    for rev in phone_soup.find_all("div", class_="user-review"):
        reviews.append(rev.text)
    # Fallback if structure changes
    price = "Unknown"
    for spec in phone_soup.find_all("td", string="Price"):
        price = spec.find_next_sibling("td").text
    specs = {li.find('span').text: li.text for li in phone_soup.find_all("li")}
    return {"reviews": reviews, "price": price, "specs": specs}

#### TOOL 2: Churn/Sentiment Scoring ####
@tool
def score_churn(reviews: list) -> float:
    """Calculates aggregate churn score from reviews using sentiment analysis.
    Args:
        reviews: List of review texts.
    Returns:
        float churn score (avg polarity; lower = less churn)
    """
    scores = [TextBlob(r).sentiment.polarity for r in reviews]
    churn_score = 1 - (sum(scores) / len(scores)) if scores else 1  # Lower is better
    return churn_score

#### TOOL 3: Find Competing Phones ####
@tool
def find_competitors(price: str) -> list:
    """Scrapes or queries GSM Arena for phones in the same price bracket.
    Args:
        price: The reference price.
    Returns:
        List of phone names.
    """
    price_value = ''.join(filter(str.isdigit, price))
    # For demo: just pick other search results in same price bracket
    # Expand with real price logic as needed
    return ["Phone1", "Phone2", "Phone3"]

#### FINAL TOOL: Recommend Phone ####
@tool
def recommend_smartphone(user_phone: str):
    """Main logic for recommendation."""
    user_data = scrape_gsmarena_reviews(user_phone)
    user_reviews = user_data['reviews']
    user_score = score_churn(user_reviews)
    user_price = user_data['price']
    competitors = find_competitors(user_price)
    best_score = user_score
    best_phone = user_phone

    for c in competitors:
        c_data = scrape_gsmarena_reviews(c)
        c_reviews = c_data['reviews']
        c_score = score_churn(c_reviews)
        if c_score < best_score:
            best_score = c_score
            best_phone = c

    # If best_phone is not user's, recommend; else try next bracket
    result = {
        "Your phone": user_phone,
        "Your phone score": user_score,
        "Best competitor": best_phone,
        "Best score": best_score,
        "Recommended": "Yes" if best_phone != user_phone else "No better phone in range",
    }
    return result

#### Gradio UI ####
def agent_ui(user_phone_name):
    """Gradio UI interface handler."""
    result = recommend_smartphone(user_phone_name)
    return f"""
    **Your phone:** {result['Your phone']}
    **Churn score:** {result['Your phone score']}
    **Best alternative:** {result['Best competitor']} (score: {result['Best score']})
    **Recommended:** {result['Recommended']}
    """

# Connect everything for Hugging Face Space
with open("prompts.yaml", 'r') as stream:
    prompt_templates = yaml.safe_load(stream)

model = HfApiModel(
    max_tokens=2096,
    temperature=0.5,
    model_id='Qwen/Qwen2.5-Coder-32B-Instruct',
)

agent = CodeAgent(
    model=model,
    tools=[recommend_smartphone],  # add your main tool, you can add others as desired
    max_steps=6,
    verbosity_level=1,
    grammar=None,
    planning_interval=None,
    name=None,
    description=None,
    prompt_templates=prompt_templates
)

GradioUI(agent_ui).launch()