frontend.py

# ============================================================
#   LEXOSYNTH - FRONTEND : GRADIO UI
#   File   : frontend.py
#
#   Provides a minimal Gradio interface that calls the Flask
#   backend at localhost:5000/analyze.
#
#   Layout :
#     - Heading at top center
#     - Text input box (rounded)
#     - Dropdown "Choose Engine" on left
#     - "Analyze" button on right
#     - Output textbox with copy button below
# ============================================================

import gradio as gr
import requests

# Flask backend URL
FLASK_URL = "http://localhost:5000"

# ============================================================
# SECTION 1 : ENGINE OPTIONS
# ============================================================

ENGINES = [
    "— Select from dropdown —",
    "Supersonic (Fast)",
    "Go Deep (Llama)",
    "Hybimix (Llama + Supersonic)",
    "Go Deep Max (Sarvam)",
    "Hybimix Max (Sarvam + Supersonic)"
]

# Maps display name to backend engine key
ENGINE_MAP = {
    "Supersonic (Fast)"              : "supersonic",
    "Go Deep (Llama)"                : "go_deep",
    "Hybimix (Llama + Supersonic)"   : "hybimix",
    "Go Deep Max (Sarvam)"           : "go_deep_max",
    "Hybimix Max (Sarvam + Supersonic)": "hybimix_max"
}

# ============================================================
# SECTION 2 : ANALYZE FUNCTION (CALLS FLASK API)
# ============================================================

def analyze(text, engine_choice):
    """
    Called when the user clicks Analyze.
    Sends text + engine to Flask /analyze and returns
    the formatted output string for display.

    Parameters:
    -----------
    text          : User input text (one or more sentences)
    engine_choice : Selected engine display name from dropdown

    Returns:
    -----------
    str : Formatted result string shown in the output textbox
    """
    # Input validation
    if not text or not text.strip():
        return "⚠  Please enter some text before clicking Analyze."

    if not engine_choice or engine_choice == "— Select from dropdown —":
        return "⚠  Please select an engine from the dropdown."

    engine_key = ENGINE_MAP.get(engine_choice)
    if not engine_key:
        return "⚠  Invalid engine selection."

    # Show which engine is running
    engine_label = engine_choice

    # AI engines can be very slow on CPU — notify user
    is_ai = engine_key in ("go_deep", "hybimix", "go_deep_max", "hybimix_max")

    try:
        response = requests.post(
            f"{FLASK_URL}/analyze",
            json    = {"text": text.strip(), "engine": engine_key},
            timeout = 600    # 10 min timeout — AI inference on CPU is slow
        )

        if response.status_code == 200:
            data = response.json()
            count  = data.get("sentence_count", 0)
            output = data.get("formatted_output", "No output received.")

            # Add a summary header
            header = (
                f"Engine : {engine_label}\n"
                f"Sentences processed : {count}\n"
                + "═" * 52 + "\n\n"
            )
            return header + output

        else:
            error = response.json().get("error", "Unknown server error.")
            return f"❌  Server error : {error}"

    except requests.exceptions.ConnectionError:
        return (
            "❌  Could not connect to the Flask backend.\n"
            "    Please make sure test.py is running."
        )
    except requests.exceptions.Timeout:
        return (
            "❌  Request timed out.\n"
            "    AI engines on CPU can take several minutes.\n"
            "    Try again or use Supersonic (Fast) for quick results."
        )
    except Exception as e:
        return f"❌  Unexpected error : {str(e)}"

# ============================================================
# SECTION 3 : CUSTOM CSS
# ============================================================

CUSTOM_CSS = """
/* Center the main heading */
.main-title {
    text-align : center;
    padding    : 24px 0 8px 0;
}

/* Rounded text input */
.rounded-input textarea,
.rounded-input input {
    border-radius : 12px !important;
}

/* Rounded output box */
.rounded-output textarea {
    border-radius : 12px !important;
    font-family   : 'Courier New', monospace;
    font-size     : 0.88rem;
}

/* Analyze button */
.analyze-btn button {
    min-height    : 46px !important;
    border-radius : 10px !important;
    font-size     : 1rem  !important;
    font-weight   : 600   !important;
}

/* Dropdown rounded */
.engine-dropdown .wrap {
    border-radius : 10px !important;
}

/* Subtle section spacing */
.section-gap {
    margin-top : 8px;
}
"""

# ============================================================
# SECTION 4 : BUILD GRADIO UI
# ============================================================

with gr.Blocks(
    title  = "Tortured Phrase Detector and Corrector",
    theme  = gr.themes.Soft(
        primary_hue   = "slate",
        secondary_hue = "blue",
        neutral_hue   = "gray"
    ),
    css    = CUSTOM_CSS
) as demo:

    # ── Heading ──
    gr.HTML("""
        <div class="main-title">
            <h1 style="
                font-size   : 2rem;
                font-weight : 700;
                color       : #1e293b;
                margin      : 0;
                letter-spacing : -0.5px;
            ">
                Tortured Phrase Detector and Corrector
            </h1>
            <p style="
                color       : #64748b;
                font-size   : 0.95rem;
                margin-top  : 6px;
            ">
                Detect and correct unnatural AIML paraphrases in scientific text
            </p>
        </div>
    """)

    # ── Input text box ──
    with gr.Column(elem_classes=["section-gap"]):
        input_text = gr.Textbox(
            placeholder  = "Paste your scientific text here. Each sentence will be analyzed individually...",
            label        = "",
            lines        = 6,
            max_lines    = 20,
            elem_classes = ["rounded-input"],
            show_label   = False
        )

        # ── Dropdown (left) + Analyze button (right) ──
        with gr.Row(equal_height=True):
            engine_dropdown = gr.Dropdown(
                choices      = ENGINES,
                value        = "— Select from dropdown —",
                label        = "Choose Engine",
                interactive  = True,
                scale        = 3,
                min_width    = 240,
                elem_classes = ["engine-dropdown"]
            )

            analyze_btn = gr.Button(
                value        = "Analyze  →",
                variant      = "primary",
                scale        = 1,
                elem_classes = ["analyze-btn"]
            )

    # ── Output textbox with built-in copy button ──
    output_text = gr.Textbox(
        label            = "Output",
        lines            = 14,
        max_lines        = 30,
        interactive      = False,
        show_copy_button = True,          # Built-in copy button
        show_label       = True,
        placeholder      = "Results will appear here after analysis...",
        elem_classes     = ["rounded-output", "section-gap"]
    )

    # ── Engine description (collapses by default) ──
    with gr.Accordion("Engine descriptions", open=False):
        gr.Markdown("""
| Engine | Speed | Description |
|--------|-------|-------------|
| **Supersonic (Fast)** | ⚡ Instant | CSV phrase lookup — no model loading needed |
| **Go Deep (Llama)** | 🐢 Slow | LLaMA 1B fine-tuned adapter |
| **Hybimix (Llama + Supersonic)** | 🐢 Slow | Both engines merged with confidence scoring |
| **Go Deep Max (Sarvam)** | 🐢 Slow | Sarvam 2B fine-tuned adapter |
| **Hybimix Max (Sarvam + Supersonic)** | 🐢 Slow | Both engines merged with confidence scoring |

> ⚠ **Note :** AI engines (Go Deep / Hybimix) run on CPU and may take **several minutes** per request.  
> Use **Supersonic (Fast)** for instant results.
        """)

    # ── Wire Analyze button to function ──
    analyze_btn.click(
        fn             = analyze,
        inputs         = [input_text, engine_dropdown],
        outputs        = output_text,
        show_progress  = "minimal"
    )

    # ── Also trigger on Enter key in text box ──
    input_text.submit(
        fn             = analyze,
        inputs         = [input_text, engine_dropdown],
        outputs        = output_text,
        show_progress  = "minimal"
    )