breakpilot-pwa/backend/ai_processor/generators/cloze.py

"""
AI Processor - Cloze Text Generator

Generate cloze (fill-in-the-blank) texts from worksheet analysis.
"""

from pathlib import Path
import json
import logging
import os
import requests

from ..config import VISION_API, BEREINIGT_DIR, get_openai_api_key

logger = logging.getLogger(__name__)

# Language codes to names
LANGUAGE_NAMES = {
    "tr": "Tuerkisch",
    "ar": "Arabisch",
    "ru": "Russisch",
    "en": "Englisch",
    "fr": "Franzoesisch",
    "es": "Spanisch",
    "pl": "Polnisch",
    "uk": "Ukrainisch",
}


def _generate_cloze_with_openai(analysis_data: dict, target_language: str = "tr") -> dict:
    """
    Generate cloze texts based on worksheet analysis.

    Important didactic requirements:
    - Multiple meaningful gaps per sentence (not just one!)
    - Difficulty level matches the original
    - Translation with the same gaps

    Args:
        analysis_data: The analysis JSON of the worksheet
        target_language: Target language for translation (default: "tr" for Turkish)

    Returns:
        Dict with cloze_items and metadata
    """
    api_key = get_openai_api_key()

    title = analysis_data.get("title") or "Arbeitsblatt"
    subject = analysis_data.get("subject") or "Allgemein"
    grade_level = analysis_data.get("grade_level") or "unbekannt"
    canonical_text = analysis_data.get("canonical_text") or ""
    printed_blocks = analysis_data.get("printed_blocks") or []

    content_parts = []
    if canonical_text:
        content_parts.append(canonical_text)
    for block in printed_blocks:
        text = block.get("text", "").strip()
        if text and text not in content_parts:
            content_parts.append(text)

    worksheet_content = "\n\n".join(content_parts)

    if not worksheet_content.strip():
        logger.warning("Kein Textinhalt fuer Lueckentext-Generierung gefunden")
        return {"cloze_items": [], "metadata": {"error": "Kein Textinhalt gefunden"}}

    target_lang_name = LANGUAGE_NAMES.get(target_language, "Tuerkisch")

    url = "https://api.openai.com/v1/chat/completions"
    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}

    system_prompt = f"""Du bist ein erfahrener Paedagoge, der Lueckentexte fuer Schueler erstellt.

WICHTIGE REGELN FUER LUECKENTEXTE:

1. MEHRERE LUECKEN PRO SATZ:
   - Erstelle IMMER mehrere sinnvolle Luecken pro Satz
   - Beispiel: "Ich habe gestern meine Hausaufgaben gemacht."
     → Luecken: "habe" UND "gemacht" (nicht nur eine!)

2. SCHWIERIGKEITSGRAD:
   - Niveau muss exakt "{grade_level}" entsprechen

3. SINNVOLLE LUECKENWOERTER:
   - Verben (konjugiert)
   - Wichtige Nomen
   - Adjektive
   - KEINE Artikel oder Praepositionen allein

4. UEBERSETZUNG:
   - Uebersetze den VOLLSTAENDIGEN Satz auf {target_lang_name}
   - Die GLEICHEN Woerter muessen als Luecken markiert sein

5. AUSGABE: Nur gueltiges JSON, kein Markdown."""

    user_prompt = f"""Erstelle Lueckentexte aus diesem Arbeitsblatt:

TITEL: {title}
FACH: {subject}
KLASSENSTUFE: {grade_level}

TEXT:
{worksheet_content}

Erstelle 5-8 Saetze mit Luecken. Gib das Ergebnis als JSON zurueck:

{{
  "cloze_items": [
    {{
      "id": "c1",
      "original_sentence": "Der vollstaendige Originalsatz ohne Luecken",
      "sentence_with_gaps": "Der Satz mit ___ fuer jede Luecke",
      "gaps": [
        {{
          "id": "g1",
          "word": "das fehlende Wort",
          "position": 0,
          "hint": "optionaler Hinweis"
        }}
      ],
      "translation": {{
        "language": "{target_language}",
        "language_name": "{target_lang_name}",
        "full_sentence": "Vollstaendige Uebersetzung",
        "sentence_with_gaps": "Uebersetzung mit ___ an gleichen Stellen"
      }}
    }}
  ],
  "metadata": {{
    "subject": "{subject}",
    "grade_level": "{grade_level}",
    "source_title": "{title}",
    "target_language": "{target_language}",
    "total_gaps": 0
  }}
}}

WICHTIG:
- Jeder Satz MUSS mindestens 2 Luecken haben!
- Position ist der Index des Wortes im Satz (0-basiert)"""

    payload = {
        "model": "gpt-4o-mini",
        "response_format": {"type": "json_object"},
        "messages": [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": user_prompt},
        ],
        "max_tokens": 3000,
        "temperature": 0.7,
    }

    response = requests.post(url, headers=headers, json=payload)
    response.raise_for_status()
    data = response.json()

    try:
        content = data["choices"][0]["message"]["content"]
        cloze_data = json.loads(content)
    except (KeyError, json.JSONDecodeError) as e:
        raise RuntimeError(f"Fehler bei Lueckentext-Generierung: {e}")

    # Calculate total number of gaps
    total_gaps = sum(len(item.get("gaps", [])) for item in cloze_data.get("cloze_items", []))
    if "metadata" in cloze_data:
        cloze_data["metadata"]["total_gaps"] = total_gaps

    return cloze_data


def _generate_cloze_with_claude(analysis_data: dict, target_language: str = "tr") -> dict:
    """Generate cloze texts with Claude API."""
    import anthropic

    api_key = os.getenv("ANTHROPIC_API_KEY")
    if not api_key:
        raise RuntimeError("ANTHROPIC_API_KEY ist nicht gesetzt.")

    client = anthropic.Anthropic(api_key=api_key)

    title = analysis_data.get("title") or "Arbeitsblatt"
    subject = analysis_data.get("subject") or "Allgemein"
    grade_level = analysis_data.get("grade_level") or "unbekannt"
    canonical_text = analysis_data.get("canonical_text") or ""
    printed_blocks = analysis_data.get("printed_blocks") or []

    content_parts = []
    if canonical_text:
        content_parts.append(canonical_text)
    for block in printed_blocks:
        text = block.get("text", "").strip()
        if text and text not in content_parts:
            content_parts.append(text)

    worksheet_content = "\n\n".join(content_parts)

    if not worksheet_content.strip():
        return {"cloze_items": [], "metadata": {"error": "Kein Textinhalt gefunden"}}

    target_lang_name = LANGUAGE_NAMES.get(target_language, "Tuerkisch")

    prompt = f"""Erstelle Lueckentexte aus diesem Arbeitsblatt.

WICHTIGE REGELN:
1. MEHRERE LUECKEN PRO SATZ (mindestens 2!)
   Beispiel: "Ich habe gestern Hausaufgaben gemacht" → Luecken: "habe" UND "gemacht"
2. Schwierigkeitsgrad: exakt "{grade_level}"
3. Uebersetzung auf {target_lang_name} mit gleichen Luecken

TITEL: {title}
FACH: {subject}
KLASSENSTUFE: {grade_level}

TEXT:
{worksheet_content}

Antworte NUR mit diesem JSON (5-8 Saetze):
{{
  "cloze_items": [
    {{
      "id": "c1",
      "original_sentence": "Vollstaendiger Satz",
      "sentence_with_gaps": "Satz mit ___ fuer Luecken",
      "gaps": [
        {{"id": "g1", "word": "Lueckenwort", "position": 0, "hint": "Hinweis"}}
      ],
      "translation": {{
        "language": "{target_language}",
        "language_name": "{target_lang_name}",
        "full_sentence": "Uebersetzung",
        "sentence_with_gaps": "Uebersetzung mit ___"
      }}
    }}
  ],
  "metadata": {{
    "subject": "{subject}",
    "grade_level": "{grade_level}",
    "source_title": "{title}",
    "target_language": "{target_language}",
    "total_gaps": 0
  }}
}}"""

    message = client.messages.create(
        model="claude-3-5-sonnet-20241022",
        max_tokens=3000,
        messages=[{"role": "user", "content": prompt}]
    )

    content = message.content[0].text

    try:
        if "```json" in content:
            content = content.split("```json")[1].split("```")[0]
        elif "```" in content:
            content = content.split("```")[1].split("```")[0]
        cloze_data = json.loads(content.strip())
    except json.JSONDecodeError as e:
        raise RuntimeError(f"Claude hat ungueltiges JSON geliefert: {e}")

    # Calculate total number of gaps
    total_gaps = sum(len(item.get("gaps", [])) for item in cloze_data.get("cloze_items", []))
    if "metadata" in cloze_data:
        cloze_data["metadata"]["total_gaps"] = total_gaps

    return cloze_data


def generate_cloze_from_analysis(analysis_path: Path, target_language: str = "tr") -> Path:
    """
    Generate cloze texts from an analysis JSON file.

    The cloze texts will:
    - Have multiple meaningful gaps per sentence
    - Match the difficulty level of the original
    - Include translation to target language

    Args:
        analysis_path: Path to *_analyse.json file
        target_language: Language code for translation (default: "tr" for Turkish)

    Returns:
        Path to generated *_cloze.json file
    """
    if not analysis_path.exists():
        raise FileNotFoundError(f"Analysedatei nicht gefunden: {analysis_path}")

    try:
        analysis_data = json.loads(analysis_path.read_text(encoding="utf-8"))
    except json.JSONDecodeError as e:
        raise RuntimeError(f"Ungueltige Analyse-JSON: {e}")

    logger.info(f"Generiere Lueckentexte fuer: {analysis_path.name}")

    # Generate cloze texts (use configured API)
    if VISION_API == "claude":
        try:
            cloze_data = _generate_cloze_with_claude(analysis_data, target_language)
        except Exception as e:
            logger.warning(f"Claude Lueckentext-Generierung fehlgeschlagen, nutze OpenAI: {e}")
            cloze_data = _generate_cloze_with_openai(analysis_data, target_language)
    else:
        cloze_data = _generate_cloze_with_openai(analysis_data, target_language)

    # Save cloze data
    out_name = analysis_path.stem.replace("_analyse", "") + "_cloze.json"
    out_path = BEREINIGT_DIR / out_name
    out_path.write_text(json.dumps(cloze_data, ensure_ascii=False, indent=2), encoding="utf-8")

    logger.info(f"Lueckentexte gespeichert: {out_path.name}")
    return out_path