AI 문서 요약: PDF/URL → 요약 리포트

[B2C]
- 학생: 논문·교재 요약
- 일반: 책·기사 요약
- 음성: 회의·인터뷰 요약

[B2B]
- 변호사: 계약서 요약
- 의료: 진료 차트 요약
- 컨설턴트: 시장 보고서
- 임원: 일일 뉴스 다이제스트

import pdfParse from 'pdf-parse';
import { generateObject } from 'ai';

async function summarizePDF(file: File) {
  const buffer = Buffer.from(await file.arrayBuffer());
  const { text, numpages } = await pdfParse(buffer);

  const result = await generateObject({
    model: anthropic('claude-sonnet-4-6'),
    schema: z.object({
      title: z.string(),
      executiveSummary: z.string().describe('3-5 sentences'),
      keyPoints: z.array(z.string()).max(7),
      actionItems: z.array(z.string()),
      quotes: z.array(z.object({
        text: z.string(),
        page: z.number().optional(),
      })),
      readingTime: z.string(),
    }),
    prompt: `Summarize this ${numpages}-page document:

${text}

Provide structured summary.`,
  });

  return result.object;
}

import { JSDOM } from 'jsdom';
import { Readability } from '@mozilla/readability';

async function summarizeURL(url: string) {
  const html = await fetch(url).then(r => r.text());
  const dom = new JSDOM(html);
  const reader = new Readability(dom.window.document);
  const article = reader.parse();

  if (!article) throw new Error('Cannot extract content');

  const result = await generateObject({
    model: anthropic('claude-sonnet-4-6'),
    schema: SummarySchema,
    prompt: `Summarize this article from ${url}:

Title: ${article.title}
${article.textContent}`,
  });

  return result.object;
}


// 또는 Browserless / ScrapingBee (JavaScript 렌더링)
const html = await fetch('https://chrome.browserless.io/content', {
  method: 'POST',
  headers: { Authorization: `Bearer ${BROWSERLESS_KEY}` },
  body: JSON.stringify({ url, waitFor: 2000 }),
}).then(r => r.text());

// Map-Reduce 패턴
async function summarizeLargeDoc(text: string) {
  // 1. Chunking
  const chunks = chunkText(text, { maxTokens: 8000 });

  // 2. Map — 각 chunk 요약
  const summaries = await Promise.all(
    chunks.map(chunk =>
      generateText({
        model: openai('gpt-4o-mini'),  // 저렴한 모델
        prompt: `Summarize:\n${chunk}`,
        maxTokens: 500,
      }).then(r => r.text)
    ),
  );

  // 3. Reduce — 요약본 통합
  const finalResult = await generateObject({
    model: anthropic('claude-sonnet-4-6'),  // 고품질 모델
    schema: SummarySchema,
    prompt: `Combine these chunk summaries into a final structured summary:

${summaries.map((s, i) => `Chunk ${i+1}:\n${s}`).join('\n\n')}`,
  });

  return finalResult.object;
}

import { openai } from '@ai-sdk/openai';

async function summarizeMeeting(audioFile: File) {
  // 1. 전사
  const transcription = await openai.audio.transcriptions.create({
    file: audioFile,
    model: 'whisper-1',
    response_format: 'verbose_json',  // timestamps 포함
  });

  // 2. 요약
  const result = await generateObject({
    model: anthropic('claude-sonnet-4-6'),
    schema: z.object({
      participants: z.array(z.string()),
      topics: z.array(z.string()),
      decisions: z.array(z.string()),
      actionItems: z.array(z.object({
        task: z.string(),
        assignee: z.string(),
        deadline: z.string().optional(),
      })),
      quotes: z.array(z.object({
        speaker: z.string(),
        text: z.string(),
        timestamp: z.string(),
      })),
    }),
    prompt: `Analyze this meeting transcript:\n${transcription.text}`,
  });

  return result.object;
}

// 이메일로 PDF 리포트
import { Resend } from 'resend';

const pdf = await generatePDF(summary);  // pdf-lib 사용

await resend.emails.send({
  from: 'Summarizer <noreply@example.com>',
  to: user.email,
  subject: `📄 요약 완료: ${summary.title}`,
  react: SummaryEmail({ summary }),
  attachments: [{ filename: 'summary.pdf', content: pdf }],
});