streamlit_app.py

import streamlit as st
import plotly.graph_objects as go
import pandas as pd
import torch

from transformers import AutoTokenizer, AutoModelForSequenceClassification

# ---------------------------------------------------
# Page Config (must be first Streamlit command)
# ---------------------------------------------------

st.set_page_config(
    page_title="Aspect-Based Intelligence",
    page_icon="✨",
    layout="wide",
    initial_sidebar_state="expanded"
)

# ---------------------------------------------------
# Load Model
# ---------------------------------------------------
MODEL_NAME = "justhariharan/Review-RAG"

@st.cache_resource
def load_model():
    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
    model.eval()
    return tokenizer, model

tokenizer, model = load_model()

labels = ["positive","negative","neutral","conflict"]

# ---------------------------------------------------
# Prediction Function
# ---------------------------------------------------

def predict_sentiment(text):

    inputs = tokenizer(
        text,
        return_tensors="pt",
        truncation=True,
        padding=True
    )

    with torch.no_grad():
        outputs = model(**inputs)

    probs = torch.softmax(outputs.logits, dim=1)[0]

    pred_id = torch.argmax(probs).item()

    return {
        "sentiment": labels[pred_id],
        "confidence": float(probs[pred_id])
    }

# ---------------------------------------------------
# UI Styling
# ---------------------------------------------------

st.markdown("""
<style>

.block-container {
padding-top:2rem;
padding-bottom:2rem;
}

.sentiment-card{
padding:1.2rem;
border-radius:12px;
margin-bottom:1rem;
box-shadow:0 4px 6px rgba(0,0,0,0.1);
color:white;
display:flex;
justify-content:space-between;
align-items:center;
}

.aspect-title{font-size:1.4rem;font-weight:700;}
.sentiment-label{text-transform:uppercase;}

.pos-theme{background:linear-gradient(135deg,#11998e,#38ef7d);}
.neg-theme{background:linear-gradient(135deg,#FF416C,#FF4B2B);}
.neu-theme{background:linear-gradient(135deg,#7F7FD5,#86A8E7);}
.con-theme{background:linear-gradient(135deg,#F2994A,#F2C94C);}

.gradient-text{
font-weight:800;
background:-webkit-linear-gradient(45deg,#4facfe,#00f2fe);
-webkit-background-clip:text;
-webkit-text-fill-color:transparent;
font-size:3rem;
}

</style>
""", unsafe_allow_html=True)

# ---------------------------------------------------
# Header
# ---------------------------------------------------

st.markdown('<h1 class="gradient-text">✨ Aspect-Based Review Intelligence</h1>', unsafe_allow_html=True)

st.markdown(
"Transforming unstructured customer feedback into real-time actionable telemetry."
)

st.divider()

# ---------------------------------------------------
# Sidebar Status
# ---------------------------------------------------

with st.sidebar:

    st.title("System Status")

    st.success("🟢 Model Loaded")
    st.info("🧠 RoBERTa Inference Ready")

    st.divider()

    st.markdown("**Tech Stack:**")
    st.markdown("`RoBERTa` `Transformers` `Streamlit`")

# ---------------------------------------------------
# Tabs
# ---------------------------------------------------

tab1, tab2, tab3 = st.tabs(
["📊 Analytics Dashboard","💬 RAG Copilot","🔍 Live Extraction"]
)

# ---------------------------------------------------
# Analytics Dashboard
# ---------------------------------------------------

with tab1:

    st.subheader("Sentiment Distribution (Demo Analytics)")

    demo_data = {
        "positive":120,
        "negative":45,
        "neutral":30,
        "conflict":15
    }

    fig = go.Figure(
        data=[go.Bar(
            x=list(demo_data.keys()),
            y=list(demo_data.values())
        )]
    )

    st.plotly_chart(fig,use_container_width=True)

# ---------------------------------------------------
# RAG Copilot (UI only demo)
# ---------------------------------------------------

with tab2:

    st.subheader("Talk to Your Product Data")

    st.info("RAG Copilot demo mode. Backend disabled in Spaces.")

    prompt = st.chat_input("Ask about product feedback")

    if prompt:

        with st.chat_message("user"):
            st.write(prompt)

        with st.chat_message("assistant"):
            st.write("RAG backend is not deployed in this demo.")

# ---------------------------------------------------
# Live Review Analysis
# ---------------------------------------------------

with tab3:

    st.subheader("NLP Extraction Sandbox")

    review_text = st.text_area(
        "Input Review:",
        height=120,
        placeholder="The laptop screen is gorgeous but the battery drains incredibly fast."
    )

    analyze_btn = st.button("🚀 Run Extraction")

    if analyze_btn and review_text:

        with st.spinner("Running RoBERTa inference..."):

            result = predict_sentiment(review_text)

            sentiment = result["sentiment"]
            conf = result["confidence"]

            theme_class="neu-theme"
            icon="⚪"

            if sentiment=="positive":
                theme_class="pos-theme"
                icon="🟢"

            elif sentiment=="negative":
                theme_class="neg-theme"
                icon="🔴"

            elif sentiment=="conflict":
                theme_class="con-theme"
                icon="🟡"

            card_html=f"""
            <div class="sentiment-card {theme_class}">
                <div>
                    <p class="aspect-title">{icon} Overall Sentiment</p>
                    <p class="sentiment-label">{sentiment}</p>
                </div>
                <div style="text-align:right;">
                    <h2>{conf*100:.0f}%</h2>
                    <p>Confidence</p>
                </div>
            </div>
            """

            st.markdown(card_html,unsafe_allow_html=True)