Adobe Core Workflow B — PDF Services

Step 2: Extract Text and Tables from PDF (Sensei AI)

// src/workflows/pdf-extract.ts
import {
  PDFServices,
  MimeType,
  ExtractPDFParams,
  ExtractElementType,
  ExtractPDFJob,
  ExtractPDFResult,
  ExtractRenditionsElementType,
} from '@adobe/pdfservices-node-sdk';
import * as fs from 'fs';
import AdmZip from 'adm-zip';

export async function extractPdfContent(
  pdfPath: string,
  options?: { tables?: boolean; figures?: boolean }
): Promise<{ text: string; tables: any[]; }> {
  const inputStream = fs.createReadStream(pdfPath);
  const inputAsset = await pdfServices.upload({
    readStream: inputStream,
    mimeType: MimeType.PDF,
  });

  const elements = [ExtractElementType.TEXT];
  if (options?.tables !== false) elements.push(ExtractElementType.TABLES);

  const params = new ExtractPDFParams({
    elementsToExtract: elements,
    ...(options?.figures && {
      elementsToExtractRenditions: [ExtractRenditionsElementType.FIGURES],
    }),
  });

  const job = new ExtractPDFJob({ inputAsset, params });
  const pollingURL = await pdfServices.submit({ job });
  const result = await pdfServices.getJobResult({
    pollingURL,
    resultType: ExtractPDFResult,
  });

  // Download and parse the result ZIP
  const resultAsset = result.result!.resource;
  const streamAsset = await pdfServices.getContent({ asset: resultAsset });
  const chunks: Buffer[] = [];
  for await (const chunk of streamAsset.readStream) {
    chunks.push(Buffer.from(chunk));
  }

  const zip = new AdmZip(Buffer.concat(chunks));
  const structuredData = JSON.parse(
    zip.readAsText('structuredData.json')
  );

  // Parse text elements
  const textElements = structuredData.elements
    .filter((el: any) => el.Text)
    .map((el: any) => el.Text);

  // Parse table elements
  const tableElements = structuredData.elements
    .filter((el: any) => el.Path?.includes('/Table'));

  return { text: textElements.join('\n'), tables: tableElements };
}

Step 3: Document Generation from Word Template

// src/workflows/pdf-docgen.ts
import {
  PDFServices,
  MimeType,
  DocumentMergeJob,
  DocumentMergeParams,
  DocumentMergeResult,
  OutputFormat,
} from '@adobe/pdfservices-node-sdk';
import * as fs from 'fs';

export async function generateDocument(
  templatePath: string,   // .docx Word template with {{tags}}
  data: Record<string, any>,
  outputPath: string,
  format: 'pdf' | 'docx' = 'pdf'
): Promise<void> {
  const inputStream = fs.createReadStream(templatePath);
  const inputAsset = await pdfServices.upload({
    readStream: inputStream,
    mimeType: MimeType.DOCX,
  });

  const params = new DocumentMergeParams({
    jsonDataForMerge: data,
    outputFormat: format === 'pdf' ? OutputFormat.PDF : OutputFormat.DOCX,
  });

  const job = new DocumentMergeJob({ inputAsset, params });
  const pollingURL = await pdfServices.submit({ job });
  const result = await pdfServices.getJobResult({
    pollingURL,
    resultType: DocumentMergeResult,
  });

  const resultAsset = result.result!.asset;
  const streamAsset = await pdfServices.getContent({ asset: resultAsset });
  const output = fs.createWriteStream(outputPath);
  streamAsset.readStream.pipe(output);

  console.log(`Document generated: ${outputPath}`);
}

// Usage: Invoice generation
// await generateDocument('./templates/invoice.docx', {
//   company: 'Acme Corp',
//   invoiceNumber: 'INV-2026-001',
//   items: [
//     { description: 'API Integration', quantity: 1, price: 5000 },
//     { description: 'Support Plan', quantity: 12, price: 200 },
//   ],
//   total: '$7,400.00',
// }, './output/invoice.pdf');

Step 4: PDF to Markdown (LLM-Friendly)

// PDF Extract API supports structured output for LLM ingestion
export async function pdfToMarkdown(pdfPath: string): Promise<string> {
  const { text } = await extractPdfContent(pdfPath, { tables: false });

  // The structuredData.json includes element paths indicating heading levels
  // For full Markdown fidelity, parse element Paths:
  //   /H1 -> # heading, /H2 -> ## heading, /L/LI -> bullet
  return text;
}

Output

• PDF files created from HTML, DOCX, or other formats
• Structured JSON with text, tables, and figures extracted from PDFs
• Dynamic documents generated from Word templates with JSON data
• Markdown text extracted from PDFs for LLM consumption

Error Handling

Resources

• PDF Services API How-Tos
• Document Generation API
• PDF Extract API
• PDF Services Node SDK Samples

Next Steps

For common errors, see adobe-common-errors.