import streamlit as st
from transformers import pipeline
import re

# Load models
emotion_classifier = pipeline("text-classification", model="bhadresh-savani/distilbert-base-uncased-emotion", top_k=1)
intent_classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
text_generator = pipeline("text2text-generation", model="google/flan-t5-base", max_new_tokens=100)

# Candidate intents
candidate_tasks = [
    "change mobile plan",
    "top up balance",
    "report service outage",
    "ask for billing support",
    "reactivate service",
    "cancel subscription",
    "check account status",
    "upgrade device"
]

# Emotion scoring
urgent_emotions = {"anger", "annoyance", "disgust", "frustration", "sadness"}
moderate_emotions = {"confusion", "concern", "nervousness", "fear"}

def get_emotion_score(emotion):
    if emotion in urgent_emotions:
        return 1.0
    elif emotion in moderate_emotions:
        return 0.6
    else:
        return 0.2

def get_content_score(text, top_intents):
    score = 0.0
    trigger_words = ["out of service", "urgent", "not working", "stopped", "can't", "immediately"]
    if any(kw in text.lower() for kw in trigger_words):
        score += 0.3
    if any(label in ["top up balance", "reactivate service", "report service outage"] for label in top_intents):
        score += 0.4
    if any(label in ["change mobile plan", "ask for billing support"] for label in top_intents):
        score += 0.2
    return min(score + 0.1, 1.0)

def generate_reply(input_text, intent):
    prompt = f"Generate a polite and helpful customer service reply to this intent: '{intent}'. Customer said: '{input_text}'"
    return text_generator(prompt)[0]['generated_text']

# UI
st.set_page_config(page_title="Customer Support Assistant", layout="centered")
st.title("📞 Smart Customer Support Assistant (for Agents Only)")

user_input = st.text_area("Enter customer's message or complaint:", height=150)

if st.button("Analyze Message"):
    if user_input.strip() == "":
        st.warning("Please enter a customer message.")
    else:
        with st.spinner("Processing..."):

            # Emotion detection
            emotion_result = emotion_classifier(user_input)
            emotion_label = emotion_result[0]['label']
            emotion_score = get_emotion_score(emotion_label)

            # Intent detection
            intent_result = intent_classifier(user_input, candidate_tasks)
            top_intents = [label for label, score in zip(intent_result['labels'], intent_result['scores']) if score > 0.2]

            # Content score
            content_score = get_content_score(user_input, top_intents)

            # Final decision score
            final_score = (0.5 * emotion_score) + (0.5 * content_score)

            st.subheader("🧾 System Summary")

            if final_score < 0.5:
                st.markdown("### 🟢 This message was handled automatically.")
                if top_intents:
                    auto_intent = top_intents[0]
                    auto_reply = generate_reply(user_input, auto_intent)
                    st.markdown("#### 🤖 Auto-Response Sent to User:")
                    st.success(auto_reply)
                else:
                    st.info("No clear intent detected. A general auto-reply was used.")
            else:
                st.markdown("### 🔴 Human Support Required")

                # Customer Profile Summary
                st.markdown("#### 👤 Customer Status:")
                st.write(f"- **Emotion detected**: {emotion_label.capitalize()}")
                st.write(f"- **Tone**: {'Urgent' if emotion_score > 0.8 else 'Concerned' if emotion_score > 0.5 else 'Calm'}")

                if top_intents:
                    st.markdown("#### 🧩 Detected Customer Needs:")
                    for intent in top_intents:
                        reply = generate_reply(user_input, intent)
                        st.markdown(f"**• {intent.capitalize()}**")
                        st.write(f"Recommended Action: {reply}")
                else:
                    st.warning("No clear intent detected. Manual review recommended.")