app.py

#!/usr/bin/env python3
"""Gradio Pangram demo app for a Hugging Face Space.

Paste a Pangram API key live, paste text, and inspect:
- overall prediction
- percentage breakdown
- annotated text spans
- flagged windows

No key or text is persisted by this app.
"""

from __future__ import annotations

import json
from collections import Counter
from dataclasses import dataclass
from typing import Any
from urllib import error, request

import gradio as gr


API_URL = "https://text.api.pangram.com/v3"

COLOR_MAP = {
    "AI": "#f1b7b0",
    "AI-assisted": "#f2d99c",
    "Human": "#d9e8d1",
    "Neutral": "#ece6d9",
}


def _coerce_int(value: Any, default: int = 0) -> int:
    try:
        return int(value)
    except (TypeError, ValueError):
        return default


def _coerce_float(value: Any) -> float | None:
    try:
        return float(value)
    except (TypeError, ValueError):
        return None


@dataclass
class Window:
    start: int
    end: int
    label: str
    confidence: float | None


def _pretty_label(value: Any) -> str:
    if value is None:
        return "-"
    text = str(value).replace("_", " ").strip()
    return " ".join(part[:1].upper() + part[1:] for part in text.split()) or "-"


def _normalize_label(value: Any) -> str:
    text = str(value or "").lower()
    if "assist" in text:
        return "AI-assisted"
    if "human" in text:
        return "Human"
    if "ai" in text:
        return "AI"
    return "Neutral"


def _post_json(url: str, api_key: str, payload: dict[str, Any]) -> dict[str, Any]:
    body = json.dumps(payload).encode("utf-8")
    req = request.Request(
        url,
        data=body,
        headers={
            "Content-Type": "application/json",
            "X-API-Key": api_key,
        },
        method="POST",
    )
    try:
        with request.urlopen(req, timeout=60) as response:
            return json.loads(response.read().decode("utf-8"))
    except error.HTTPError as exc:
        raw = exc.read().decode("utf-8", errors="replace")
        try:
            data = json.loads(raw)
        except json.JSONDecodeError:
            data = {"error": raw or exc.reason}
        message = data.get("error") or data.get("detail") or data.get("message") or str(exc)
        raise gr.Error(f"Pangram request failed: {message}") from exc
    except Exception as exc:  # pragma: no cover - network edge cases
        raise gr.Error(f"Pangram request failed: {exc}") from exc


def _extract_windows(raw_windows: Any) -> list[Window]:
    windows: list[Window] = []
    if not isinstance(raw_windows, list):
        return windows

    for item in raw_windows:
        if not isinstance(item, dict):
            continue
        start = max(0, _coerce_int(item.get("start_index", 0), 0))
        end = max(start, _coerce_int(item.get("end_index", start), start))
        confidence = _coerce_float(item.get("confidence"))
        windows.append(
            Window(
                start=start,
                end=end,
                label=_normalize_label(item.get("label")),
                confidence=confidence,
            )
        )
    windows.sort(key=lambda w: (w.start, w.end))
    return windows


def _build_highlighted_segments(text: str, windows: list[Window]) -> list[tuple[str, str | None]]:
    if not text:
        return []

    segments: list[tuple[str, str | None]] = []
    cursor = 0

    for window in windows:
        start = min(max(window.start, cursor, 0), len(text))
        end = min(max(window.end, start), len(text))

        if start > cursor:
            segments.append((text[cursor:start], None))

        if end > start:
            segments.append((text[start:end], window.label))
            cursor = end

    if cursor < len(text):
        segments.append((text[cursor:], None))

    if not segments:
        segments.append((text, None))

    return segments


def _percentages_table(raw_percentages: Any) -> list[list[str]]:
    if not isinstance(raw_percentages, dict):
        return []

    rows: list[list[str]] = []
    sortable: list[tuple[str, float]] = []
    for label, value in raw_percentages.items():
        pct = _coerce_float(value)
        if pct is None:
            continue
        sortable.append((str(label), pct))

    for label, pct in sorted(sortable, key=lambda item: item[1], reverse=True):
        rows.append([_pretty_label(label), f"{pct:.1f}%"])
    return rows


def _flagged_windows_table(text: str, windows: list[Window]) -> list[list[str]]:
    rows: list[list[str]] = []
    for index, window in enumerate(windows, start=1):
        if window.label not in {"AI", "AI-assisted"}:
            continue
        excerpt = text[window.start:window.end].strip() or "[empty window]"
        confidence = "-" if window.confidence is None else f"{window.confidence:.3f}"
        rows.append([str(index), window.label, confidence, excerpt])
    return rows


def _summary_markdown(result: dict[str, Any], windows: list[Window]) -> str:
    prediction = _pretty_label(result.get("prediction_short") or result.get("prediction"))
    public_link = result.get("public_dashboard_link")
    flagged = sum(1 for window in windows if window.label in {"AI", "AI-assisted"})
    counts = Counter(window.label for window in windows)
    top_label = counts.most_common(1)[0][0] if counts else "-"

    lines = [
        f"**Prediction:** {prediction}",
        f"**Flagged windows:** {flagged} / {len(windows)}",
        f"**Most frequent window label:** {top_label}",
    ]
    if public_link:
        lines.append(f"**Public dashboard:** {public_link}")
    return "\n\n".join(lines)


def analyze_text(api_key: str, text: str, public_dashboard_link: bool) -> tuple[Any, ...]:
    api_key = (api_key or "").strip()
    text = text or ""

    if not api_key:
        raise gr.Error("Paste a Pangram API key first.")
    if not text.strip():
        raise gr.Error("Paste text to analyze first.")

    payload = {
        "text": text,
        "public_dashboard_link": bool(public_dashboard_link),
    }
    result = _post_json(API_URL, api_key, payload)
    windows = _extract_windows(result.get("windows"))
    highlighted = _build_highlighted_segments(text, windows)
    percentages = _percentages_table(result.get("percentages"))
    flagged = _flagged_windows_table(text, windows)
    summary = _summary_markdown(result, windows)

    return (
        summary,
        highlighted,
        percentages,
        flagged,
        result,
    )


def clear_outputs() -> tuple[Any, ...]:
    return ("", "", False, "", [], [], {}, [])


with gr.Blocks(title="Pangram Live Demo") as demo:
    gr.Markdown(
        """
        # Pangram Live Demo

        Paste a Pangram API key and a text sample to inspect flagged spans live.
        The key is used only for the current request and is not stored by the app.
        """
    )

    with gr.Row():
        with gr.Column(scale=4):
            api_key = gr.Textbox(
                label="Pangram API key",
                type="password",
                placeholder="Paste live API key here",
            )
            source_text = gr.Textbox(
                label="Text to analyze",
                lines=16,
                placeholder="Paste text here...",
            )
            public_link = gr.Checkbox(
                label="Request public dashboard link",
                value=False,
            )
            with gr.Row():
                analyze = gr.Button("Analyze", variant="primary")
                clear = gr.Button("Clear")

        with gr.Column(scale=5):
            summary = gr.Markdown(label="Summary")
            annotated = gr.HighlightedText(
                label="Annotated output",
                color_map=COLOR_MAP,
                combine_adjacent=True,
                show_legend=True,
                show_inline_category=False,
            )
            percentages = gr.Dataframe(
                headers=["Label", "Percentage"],
                datatype=["str", "str"],
                row_count=(0, "dynamic"),
                col_count=(2, "fixed"),
                label="Percentages",
                wrap=True,
            )
            flagged = gr.Dataframe(
                headers=["#", "Label", "Confidence", "Excerpt"],
                datatype=["str", "str", "str", "str"],
                row_count=(0, "dynamic"),
                col_count=(4, "fixed"),
                label="Flagged windows",
                wrap=True,
            )
            raw_json = gr.JSON(label="Raw Pangram response")

    analyze.click(
        fn=analyze_text,
        inputs=[api_key, source_text, public_link],
        outputs=[summary, annotated, percentages, flagged, raw_json],
        api_name=False,
    )

    clear.click(
        fn=clear_outputs,
        inputs=[],
        outputs=[api_key, source_text, public_link, summary, percentages, flagged, raw_json, annotated],
        api_name=False,
    )


if __name__ == "__main__":
    demo.launch()