extractor.py

from typing import Dict, Any


def analyze_context(message: str) -> Dict[str, Any]:
    # Text normalisieren und für Wort-Splitting vorbereiten
    text_raw = message.lower()
    
    # Satzzeichen entfernen für exaktes Word-Matching
    clean_text = text_raw
    for char in [".", ",", "!", "?", ";", ":", "-", "_", "(", ")", "[", "]", "\n", "\r"]:
        clean_text = clean_text.replace(char, " ")
        
    # In einzelne Wörter splitten, um Sub-String-Fehler (z.B. "ich" in "sicherlich") zu vermeiden
    words_set = set(clean_text.split())

    scores = {
        "Business": 0.0,
        "Technik": 0.0,
        "Kreativ": 0.0,
        "Persönlich": 0.0
    }

    business_words = [
        "business", "umsatz", "kunden", "angebot", "markt", "preis",
        "verkauf", "agentur", "strategie", "monetarisierung", "framework"
    ]

    tech_words = [
        "code", "api", "router", "fastapi", "gradio", "python",
        "github", "hugging face", "hf", "mcp", "server", "datenbank"
    ]

    creative_words = [
        "bild", "video", "content", "design", "persona", "prompt",
        "reel", "shorts", "style", "look", "story"
    ]

    personal_words = [
        "ich", "mein", "problem", "hilfe", "stress", "plan",
        "ziel", "entscheidung", "gedanke"
    ]

    # Abgleich auf echte Wort-Treffer oder exakte Phrasen
    for word in business_words:
        if word in words_set or (" " in word and word in text_raw):
            scores["Business"] += 1

    for word in tech_words:
        if word in words_set or (" " in word and word in text_raw):
            scores["Technik"] += 1

    for word in creative_words:
        if word in words_set or (" " in word and word in text_raw):
            scores["Kreativ"] += 1

    for word in personal_words:
        if word in words_set:
            scores["Persönlich"] += 0.5

    total = sum(scores.values()) or 1

    distribution = {
        key: round(value / total, 2)
        for key, value in scores.items()
    }

    raw_primary = max(distribution, key=distribution.get)
    actual_confidence = distribution.get(raw_primary, 0.5)

    # Wenn der stärkste Intent unter 40% liegt, deklarieren wir es als gemischten Kontext
    if distribution[raw_primary] < 0.4:
        primary = "Mixed"
    else:
        primary = raw_primary

    intent = detect_intent(text_raw)
    complexity = detect_complexity(text_raw)

    return {
        "primary_context": {
            "type": primary,
            "confidence": actual_confidence if primary != "Mixed" else round(actual_confidence, 2)
        },
        "context_distribution": distribution,
        "intent_layer": intent,
        "complexity_level": complexity,
        "routing_hint": build_routing_hint(primary, intent, complexity),
        "reasoning": "Kontext wurde anhand exakter Schlüsselwörter und Signalgruppen bestimmt."
    }


def detect_intent(text: str) -> Dict[str, Any]:
    if any(w in text for w in ["bauen", "erstellen", "setup", "einbauen", "implementieren"]):
        return {"type": "Aufbau", "confidence": 0.85}

    if any(w in text for w in ["analysieren", "bewerten", "prüfen", "vergleich"]):
        return {"type": "Analyse", "confidence": 0.8}

    if any(w in text for w in ["idee", "konzept", "vision"]):
        return {"type": "Idee", "confidence": 0.75}

    if any(w in text for w in ["problem", "fehler", "geht nicht", "kaputt"]):
        return {"type": "Problem", "confidence": 0.85}

    if any(w in text for w in ["plan", "nächste schritte", "roadmap"]):
        return {"type": "Planung", "confidence": 0.8}

    return {"type": "Umsetzung", "confidence": 0.6}


def detect_complexity(text: str) -> str:
    length = len(text)

    if length > 1500:
        return "high"
    if length > 500:
        return "medium"
    return "low"


def build_routing_hint(primary: str, intent: Dict[str, Any], complexity: str) -> Dict[str, Any]:
    if primary == "Technik":
        target = "DeepSeek"
    elif primary == "Business":
        target = "Hermes"
    elif primary == "Kreativ":
        target = "JoyAI"
    elif primary == "Persönlich":
        target = "Hermes"
    else:
        target = "Orchestrator"

    priority = "high" if complexity == "high" else "medium"

    return {
        "target_system": target,
        "priority": priority,
        "intent": intent["type"]
    }