app.py

import os
import re
import requests
import torch
import streamlit as st
from langchain_huggingface import HuggingFaceEndpoint
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from transformers import pipeline
from langdetect import detect  # Ensure this package is installed

# ✅ Check for GPU or Default to CPU
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"✅ Using device: {device}")  # Debugging info

# ✅ Environment Variables
HF_TOKEN = os.getenv("HF_TOKEN")
if HF_TOKEN is None:
    raise ValueError("HF_TOKEN is not set. Please add it to your environment variables.")

NASA_API_KEY = os.getenv("NASA_API_KEY")
if NASA_API_KEY is None:
    raise ValueError("NASA_API_KEY is not set. Please add it to your environment variables.")

# ✅ Set Up Streamlit
st.set_page_config(page_title="HAL - NASA ChatBot", page_icon="🚀")

# ✅ Initialize Session State Variables (Ensuring Chat History Persists)
if "chat_history" not in st.session_state:
    st.session_state.chat_history = [{"role": "assistant", "content": "Hello! How can I assist you today?"}]
if "response_ready" not in st.session_state:
    st.session_state.response_ready = False
if "follow_up" not in st.session_state:
    st.session_state.follow_up = ""

# ✅ Initialize Hugging Face Model (Explicitly Set to CPU/GPU)
def get_llm_hf_inference(model_id="mistralai/Mistral-7B-Instruct-v0.3", max_new_tokens=512, temperature=0.7):
    return HuggingFaceEndpoint(
        repo_id=model_id,
        max_new_tokens=max_new_tokens,
        temperature=temperature,
        token=HF_TOKEN,
        task="text-generation",
        device=-1 if device == "cpu" else 0  # ✅ Force CPU (-1) or GPU (0)
    )

# ✅ NASA API Function
def get_nasa_apod():
    url = f"https://api.nasa.gov/planetary/apod?api_key={NASA_API_KEY}"
    response = requests.get(url)
    if response.status_code == 200:
        data = response.json()
        return data.get("url", ""), data.get("title", ""), data.get("explanation", "")
    return "", "NASA Data Unavailable", "I couldn't fetch data from NASA right now."

# ✅ Sentiment Analysis (Now Uses Explicit Device)
sentiment_analyzer = pipeline(
    "sentiment-analysis",
    model="distilbert/distilbert-base-uncased-finetuned-sst-2-english",
    device=-1 if device == "cpu" else 0  # ✅ Force CPU (-1) or GPU (0)
)

def analyze_sentiment(user_text):
    result = sentiment_analyzer(user_text)[0]
    return result['label']

# ✅ Intent Detection
def predict_action(user_text):
    if "NASA" in user_text.lower() or "space" in user_text.lower():
        return "nasa_info"
    return "general_query"

# ✅ Ensure English Responses
def ensure_english(text):
    try:
        detected_lang = detect(text)
        if detected_lang != "en":
            return "⚠️ Sorry, I only respond in English. Can you rephrase your question?"
    except:
        return "⚠️ Language detection failed. Please ask your question again."
    return text

# ✅ Follow-Up Question Generation
def generate_follow_up(user_text):
    """Generates a structured follow-up question in a concise format."""
    
    prompt_text = (
        f"Given the user's question: '{user_text}', generate a SHORT follow-up question in the format: "
        "'Would you like to learn more about [related topic] or explore something else?'. "
        "Ensure it's concise and structured exactly as requested without extra commentary."
    )

    hf = get_llm_hf_inference(max_new_tokens=30, temperature=0.6)  # 🔥 Lower temp for consistency
    output = hf.invoke(input=prompt_text).strip()

    # ✅ Extract the relevant part using regex to remove unwanted symbols or truncations
    cleaned_output = re.sub(r"```|''|\"", "", output).strip()

    # ✅ Ensure output is formatted correctly
    if "Would you like to learn more about" not in cleaned_output:
        cleaned_output = "Would you like to explore another related topic or ask about something else?"

    return cleaned_output

# ✅ Main Response Function
def get_response(system_message, chat_history, user_text, max_new_tokens=512):
    action = predict_action(user_text)

    # ✅ Handle NASA-Specific Queries
    if action == "nasa_info":
        nasa_url, nasa_title, nasa_explanation = get_nasa_apod()
        response = f"**{nasa_title}**\n\n{nasa_explanation}"
        follow_up = generate_follow_up(user_text)
        chat_history.extend([
            {'role': 'user', 'content': user_text},
            {'role': 'assistant', 'content': response},
            {'role': 'assistant', 'content': follow_up}
        ])
        return response, follow_up, chat_history, nasa_url

    # ✅ Invoke Hugging Face Model
    hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.9)

    filtered_history = "\n".join(f"{msg['role']}: {msg['content']}" for msg in chat_history)

    prompt = PromptTemplate.from_template(
        "[INST] {system_message}\n\nCurrent Conversation:\n{chat_history}\n\n"
        "User: {user_text}.\n [/INST]\n"
        "AI: Provide a detailed explanation. Use a conversational tone. "
        "🚨 Answer **only in English**."
        "\nHAL:"
    )

    chat = prompt | hf.bind(skip_prompt=True) | StrOutputParser(output_key='content')
    response = chat.invoke(input=dict(system_message=system_message, user_text=user_text, chat_history=filtered_history))
    response = response.split("HAL:")[-1].strip() if "HAL:" in response else response.strip()

    response = ensure_english(response)

    if not response:
        response = "I'm sorry, but I couldn't generate a response. Can you rephrase your question?"

    follow_up = generate_follow_up(user_text)

    chat_history.extend([
        {'role': 'user', 'content': user_text},
        {'role': 'assistant', 'content': response},
        {'role': 'assistant', 'content': follow_up}
    ])

    return response, follow_up, chat_history, None

# ✅ Streamlit UI
st.title("🚀 HAL - NASA AI Assistant")

# ✅ Justify all chatbot responses
st.markdown("""
    <style>
    .user-msg, .assistant-msg {
        padding: 10px;
        border-radius: 10px;
        margin-bottom: 5px;
        width: fit-content;
        max-width: 80%;
        text-align: justify;
    }
    .user-msg { background-color: #696969; color: white; }
    .assistant-msg { background-color: #333333; color: white; }
    .container { display: flex; flex-direction: column; align-items: flex-start; }
    @media (max-width: 600px) { .user-msg, .assistant-msg { font-size: 16px; max-width: 100%; } }
    </style>
""", unsafe_allow_html=True)

# ✅ Reset Chat Button
if st.sidebar.button("Reset Chat"):
    st.session_state.chat_history = [{"role": "assistant", "content": "Hello! How can I assist you today?"}]
    st.session_state.response_ready = False
    st.session_state.follow_up = ""

# ✅ Chat UI
user_input = st.chat_input("Type your message here...")

if user_input:
    response, follow_up, st.session_state.chat_history, image_url = get_response(
        system_message="You are a helpful AI assistant.",
        user_text=user_input,
        chat_history=st.session_state.chat_history
    )

    if response:
        st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {response}</div>", unsafe_allow_html=True)

    if follow_up:
        st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {follow_up}</div>", unsafe_allow_html=True)

    if image_url:
        st.image(image_url, caption="NASA Image of the Day")

    st.session_state.response_ready = True

if st.session_state.response_ready and st.session_state.follow_up:
    st.markdown(f"<div class='assistant-msg'><strong>HAL:</strong> {st.session_state.follow_up}</div>", unsafe_allow_html=True)
    st.session_state.response_ready = False