keywords.py

REASONING_KEYWORDS = [
    # explicit reasoning requests
    "prove", "demonstrate", "derive", "justify", "verify",
    "show that", "walk through", "step by step", "reason through",
    "chain of reasoning", "rigorous", "formal proof",

    # analysis/comparison
    "analyze", "analysis of", "compare and contrast",
    "evaluate", "critically assess", "explain why",
    "explain how", "what causes", "implications of",

    # problem solving
    "solve", "solution to", "how would you approach",
    "strategy for", "optimize", "algorithm for",

    # technical domains
    "theorem", "lemma", "corollary",
    "complexity analysis", "big o", "time complexity",
    "mathematical", "statistical", "probabilistic",
    "model the", "simulate",
]

CODE_KEYWORDS = [
    "await", "async", "print(", "console.log(",
    "code", ".ts", ".js", ".py", ".repy", ".rb",
    "gnu", "gcc", "clang", "clang++", "program",
    "coding"
]

CREATIVE_KEYWORDS = [
    # cinematic cues
    "cinematic", "film still", "movie scene",
    "epic", "dramatic lighting", "moody lighting",
    "volumetric lighting", "depth of field",
    "anamorphic lens", "8k", "4k",

    # art styles
    "concept art", "digital painting",
    "fantasy art", "sci-fi", "mythical",
    "cyberpunk", "steampunk",
    "baroque", "surreal", "abstract",
    "oil painting", "watercolor",

    # rendering engines
    "octane render", "unreal engine",
    "ray tracing", "global illumination",

    # emotional narrative framing
    "emotional portrait", "story scene",
    "hero shot", "dramatic pose",
]

STRUCTURED_KEYWORDS = [
    "return as json",
    "output json",
    "json schema",
    "format as json",
    "structured output",
    "extract entities",
    "extract fields",
    "parse this",
    "convert to table",
    "create a table",
    "categorize into",
    "classify",
    "label the following",
    "taxonomy",
    "generate schema",
]

MATH_PATTERNS = [
    r"\b∫\b", r"\b∑\b", r"\b∂\b",
    r"\bmatrix\b",
    r"\blimit\b",
    r"\bintegral\b",
    r"\bderivative\b",
    r"\bdifferential equation\b",
    r"\blinear algebra\b",
    r"\boptimi[sz]e\b",
    r"\bgradient\b",
    r"\bbackprop\b",
    r"\bproof\b",
    r"\btheorem\b",
]

LIGHTWEIGHT_KEYWORDS = [
    "hello", "hi", "hey",
    "thanks", "thank you",
    "define", "definition of",
    "what is", "who is",
    "quick question",
    "short answer",
    "brief explanation",
    "summarize",
    "paraphrase",
    "rewrite this",
]

def is_long_context(messages: list) -> bool:
    total_chars = sum(len(m.get("content", "")) for m in messages)
    return total_chars > 4000


def contains_code(prompt: str) -> bool:
    if "```" in prompt:
        return True
    for kw in CODE_KEYWORDS:
        if kw in prompt:
            return True
    return False

def is_code_heavy(prompt: str, code_present: bool, long_context: bool) -> bool:
    """
    Determines whether the coding task is substantial enough
    to require a code-optimized or larger model.
    """

    if not code_present:
        return False

    heavy_patterns = [
        r"\brefactor\b",
        r"\boptimi[sz]e\b",
        r"\bdebug\b",
        r"\bfix this\b",
        r"\barchitecture\b",
        r"\bdesign pattern\b",
        r"\bscalable\b",
        r"\bmicroservice\b",
        r"\bmultiple files\b",
        r"\bentire project\b",
        r"\bcodebase\b",
        r"\bperformance\b",
    ]

    for pattern in heavy_patterns:
        if re.search(pattern, prompt):
            return True

    if prompt.count("```") >= 2:
        return True

    if long_context:
        return True

    return False

def is_math_heavy(prompt: str) -> bool:
    for pattern in MATH_PATTERNS:
        if re.search(pattern, prompt):
            return True
    return False

def is_structured_task(prompt: str) -> bool:
    for kw in STRUCTURED_KEYWORDS:
        if kw in prompt:
            return True
    return False

def multiple_questions(prompt: str) -> bool:
    return prompt.count("?") >= 3