linkedin insight
How to Split a Multi-Document PDF Using JavaScript and Google Cloud Document AI

How to Split a Multi-Document PDF Using JavaScript and Google Cloud Document AI

Software Development
Oct 11, 2024
6-7 min

Share blog

Introduction

In this tutorial, I will guide you through a process of splitting a PDF that contains multiple documents using JavaScript, Google Cloud’s Document AI, and the pdf-lib library. This feature is useful when you have a PDF with several documents, each identified by page numbers (e.g., “Page 1 of 2” for the first document, “Page 1 of 3” for the second document, etc.). Document AI will help extract page number data, and then we’ll split the PDF accordingly.

Step 1: Understanding the Problem

Consider a PDF with multiple documents, each identified by page numbers:

  • The first document has 2 pages, labeled “Page 1 of 2”, “Page 2 of 2”.

The second document has 3 pages, labeled “Page 1 of 3”, “Page 2 of 3”, “Page 3 of 3”. We’ll use OCR (Optical Character Recognition) to extract these page numbers and split the PDF into separate files for each document.

Image

Step 2: Setting Up Google Cloud Document AI

To OCR the page numbers, we will use Google Cloud Document AI’s Custom Extractor.

1. Create a Google Cloud Account if you don’t have one.

2. Set up Document AI by searching for it in the GCP Console

Image

3. Create a Custom Processor by selecting the Custom Extractor model.

Image

4. Select Custom extractor as our processor.

Image

5. Upload Training Documents: Upload sample PDFs to train our processor .

Image

6. Create Labels: Annotate the page numbers and total page count fields, creating two labels: page_no and page_total. For optimal accuracy, label at least 100 pages across 20 documents.

Image
Image

7. Train and Deploy the model.

Step 3: Extracting Page Numbers from the PDF Using Document AI

Once the processor is trained and deployed, you can extract labeled data like page numbers and total pages from the PDF. Here’s how we do it in JavaScript:

typescript
1const name = `projects/${projectId}/locations/${location}/processors/${processorId}`;
2const buffer = await getTheArrayBufferFromPdfUrl(s3Url);
3const encodedImage = Buffer.from(buffer).toString('base64');
4
5const request = {
6 name,
7 rawDocument: {
8 content: encodedImage,
9 mimeType: 'application/pdf',
10 },
11};
12
13const [result] = await client.processDocument(request);
14const { document } = result;
15const { entities } = document;
16const pages = formatData(entities);
17const pagesToSplit = getPdfPagesToSplit(pages);

This function organizes the extracted data into a structured array containing each page’s number and total page count.

Step 4: Identifying Document Boundaries

We then determine the starting and ending pages for each document inside the PDF:

javascript
1getPdfPagesToSplit = (pages) => {
2 const pdfPages = [];
3 let count = 0;
4 let skipCount = 0;
5
6 for (const page of pages) {
7 count++;
8 if (skipCount) {
9 skipCount--;
10 continue;
11 }
12
13 if (page.page_total == 1) {
14 pdfPages.push({ number: +page.number + 1, start: count, end: count });
15 } else if (page.page_total > 1) {
16 skipCount = page.page_total - 1;
17 pdfPages.push({ number: +page.number + 1, start: count, end: count + +page.page_total - 1 });
18 }
19 }
20
21 return pdfPages;
22};
23
24 },
25};
26
27const [result] = await client.processDocument(request);
28const {document} = result;
29const {entities} = document;
30const pages = formatData(entities);
31const pagesToSplit = getPdfPagesToSplit(pages);

Step 5: Splitting the PDF Using pdf-lib

Once we have the start and end pages, we can split the PDF using pdf-lib:

javascript
1extractPdfPage = async (arrayBuff, pageToSplit) => {
2 const pdfSrcDoc = await PDFDocument.load(arrayBuff);
3 const pdfNewDoc = await PDFDocument.create();
4 const pages = await pdfNewDoc.copyPages(pdfSrcDoc, range(pageToSplit.start, pageToSplit.end));
5 pages.forEach(page => pdfNewDoc.addPage(page));
6
7 const newPdf = await pdfNewDoc.save();
8 return newPdf;
9};

Here, pdf-lib copies and saves the pages of each document as a new PDF.

Step 6: Upload or Download the Split PDFs

Now, we can take the split PDFs from SplittedPdfs and either upload them to a cloud service or download them to the user’s machine:

javascript
1const SplittedPdfs = [];
2for (const pageToSplit of pagesToSplit) {
3 const splittedPdf = await extractPdfPage(imageFile, pageToSplit);
4 SplittedPdfs.push(splittedPdf);
5}
6// Now you can use SplittedPdfs as per your needs.

Conclusion

This tutorial demonstrates how to split a multi-document PDF using JavaScript, Document AI, and pdf-lib. We covered setting up Document AI, extracting page numbers, and splitting the PDF based on those page numbers. With these steps, you can easily implement this feature in your own applications.

Blogs

Discover the latest insights and trends in technology with the Omax Tech Blog.

View All Blogs
DynamoDB multi-tenant architecture for secure data isolation.
6-10 min
August 13, 2026

Multi-Tenancy Patterns in DynamoDB: Silo, Pool, and Bridge Models

If you've already made the jump from a relational database to DynamoDB see our guide on moving relational data from SQL to DynamoDB...

Read More
Cursor IDE generating a Figma design draft from a Jira ticket - visualizing AI‑assisted design workflow.
8-10 min
August 10, 2026

We stopped leaving the IDE to design. Here’s our Cursor → Figma flow

Cursor drafts fast, catches gaps early, and still clips fields and breaks layouts. Here's the real pros-and-cons breakdown of our workflow...

Read More
AWS DevOps Agent automating AI-powered on-call incident response
6-8 min
August 07, 2026

AWS DevOps Agent: How AI is Automating On-Call Incident Response

If you've ever been on call during a production outage, you know how stressful it can be. Alerts start firing, dashboards light up, and suddenly you're jumping between monitoring tools...

Read More
Next.js pre-build script for detecting missing images before deployment
6-10 min
August 06, 2026

Catch Missing Images Before Deploy: A Simple Pre-Build Script for Next.js

How Omax Tech added a lightweight image validation gate to Next.js 15 builds on Vercel...

Read More
Storybook MCP + Amazon Bedrock + Strands: Teaching your LLM to build UI from your real design system catalog
10-15 min
August 04, 2026

Teach your LLM your design system: Storybook MCP + Amazon Bedrock + Strands

How to stop models inventing buttons and make them build UI from your real component catalog. Most "AI UI" demos look great until you paste the markup into a real product. The fix is not a smarter prompt...

Read More
Configure Self Hosted GitLab Repository Mirroring
6-8 min
July 30, 2026

Configure Self Hosted GitLab Repository Mirroring

Self hosted GitLab Repository Mirroring is a powerful feature that automatically synchronizes repositories between GitLab and external Git providers...

Read More
Illustration of Event Sourcing concepts for scalable software architecture and distributed systems.
4-8 min
July 20, 2026

Event Sourcing: A Foundation Guide

Event Sourcing is an architectural pattern where every state change is recorded as an immutable event rather than updating a database row in place...

Read More
AWS cloud security best practices with developer coding environment and cloud technology infrastructure
6-10 min
July 15, 2026

AWS Security Best Practices Every Business Should Follow

As more organizations migrate their applications and critical workloads to AWS, securing cloud environments has become a business priority rather than just an IT responsibility...

Read More
Futuristic cloud computing illustration with glowing data and AI-powered server floating in a digital neon environment.
6-10 min
June 22, 2026

AWS Migration Checklist: A Practical Roadmap for Modern Businesses

Migrating businesses to AWS offers many benefits, including cost optimization, improved security, and greater scalability. However, a successful migration requires careful planning and execution. Otherwise, organizations may experience...

Read More

Ready to Work With Us?

Most engagements start with a 20-minute conversation. No pitch, no pressure - just an honest discussion about what you're building and whether we're the right fit.