Writesonic AI写作工具架构

from typing import List, Dict, Optionalimport hashlib class WritesonicPipeline:    TEMPLATES = {        "blog": "Write a blog post about {topic} for {audience}",        "product": "Write product description for {product} with features: {features}",        "ad": "Create {platform} ad for {product} targeting {audience}",    }    LANG_PROMPTS = {        "en": {"style": "concise", "structure": "intro-body-conclusion"},        "zh": {"style": "four_char_idioms", "structure": "total-part-total"},        "ja": {"style": "polite_keigo", "structure": "intro-body-conclusion"},    }     def __init__(self):        self.cache = {}        self.quality_scores = []     def detect_language(self, text: str) -> str:        zh_count = sum(1 for c in text if '\u4e00' <= c <= '\u9fff')        ja_markers = sum(1 for c in text if c in '\u3040\u3041\u3042\u3043\u3044\u3045')        if ja_markers > 3: return "ja"        if zh_count > len(text) * 0.3: return "zh"        return "en"     def fill_template(self, template_key: str, variables: Dict) -> str:        template = self.TEMPLATES.get(template_key, "")        for k, v in variables.items():            template = template.replace("{" + k + "}", str(v))        return template     def generate_outline(self, topic: str, lang: str) -> str:        cache_key = hashlib.md5(f"outline:{topic}:{lang}".encode()).hexdigest()        if cache_key in self.cache: return self.cache[cache_key]        style = self.LANG_PROMPTS.get(lang, self.LANG_PROMPTS["en"])        outline = f"Outline[{lang}]: 1.Intro-{topic} 2.Body 3.Conclusion (style:{style['style']})"        self.cache[cache_key] = outline        return outline     def expand_sections(self, outline: str, detail_level: int) -> str:        sections = outline.split(". ")        expanded = []        for s in sections:            if s.strip(): expanded.append(f"[{s}] -> {detail_level} paragraphs")        return "\n".join(expanded)     def polish_content(self, content: str, lang: str) -> str:        style = self.LANG_PROMPTS.get(lang, self.LANG_PROMPTS["en"])        return f"Polished({style['style']}): {content[:80]}..."     def fact_check(self, content: str) -> Dict:        claims = ["stat1", "date1", "name1"]        verified = sum(1 for c in claims if hash(c) % 3 != 0)        return {"total_claims": len(claims), "verified": verified, "flagged": len(claims) - verified}     def plagiarism_check(self, content: str) -> Dict:        content_hash = hashlib.md5(content.encode()).hexdigest()        originality = 95 + (int(content_hash[:2], 16) % 5)        return {"originality_pct": originality, "passed": originality >= 95}     def run_pipeline(self, template_key: str, variables: Dict, lang: str = "en") -> Dict:        prompt = self.fill_template(template_key, variables)        outline = self.generate_outline(variables.get("topic", "general"), lang)        draft = self.expand_sections(outline, detail_level=3)        polished = self.polish_content(draft, lang)        fc = self.fact_check(polished)        pl = self.plagiarism_check(polished)        score = (fc["verified"] / max(fc["total_claims"], 1)) * 50 + (pl["originality_pct"] / 100) * 50        return {"outline": outline, "content": polished, "fact_check": fc, "plagiarism": pl, "quality_score": round(score, 1)} pipeline = WritesonicPipeline()result = pipeline.run_pipeline("blog", {"topic": "AI Writing", "audience": "marketers"}, lang="en")print(f"Quality: {result['quality_score']}")print(f"Originality: {result['plagiarism']['originality_pct']}%")result_zh = pipeline.run_pipeline("product", {"product": "Smart Watch", "features": "heart rate, GPS"}, lang="zh")print(f"Chinese Quality: {result_zh['quality_score']}")