Sanidhya

Posted on Apr 25

I built an AI-powered PDF generation API — here's how

#webdev #javascript #ai #api

PDF generation from code is still painful in 2026. You either wrestle with complex libraries that need 200+ lines for a simple invoice, or pay for bloated enterprise services.

So I built PDFGen AI — a simple REST API where you send HTML and get a PDF URL back. Or better — describe what you want in plain English and AI generates the template for you.

The Problem

Every developer who's tried to generate PDFs programmatically knows the pain:

wkhtmltopdf — outdated, rendering issues, painful to install on servers
Puppeteer/Playwright — powerful but heavy, needs headless Chrome
jsPDF — client-side only, limited styling
PDFKit — low-level, you're drawing rectangles manually
Paid services — $50-200/month for what should be a simple API call

All I wanted was: send HTML, get a PDF. That's it.

The Solution: One API Call

curl -X POST https://pdfgen-api.vercel.app/api/generate \
  -H "Authorization: Bearer pk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"html": "<h1>Invoice #001</h1><p>Amount: $500</p>"}'

Response:

{
  "success": true,
  "url": "https://storage.supabase.co/pdfs/invoice-abc123.pdf"
}

That's the entire integration. No SDKs. No config files. No dependencies.

The AI Magic

Instead of writing HTML yourself, you can use AI to do the heavy lifting.

Generate a Template from a Description

curl -X POST https://pdfgen-api.vercel.app/api/ai/template \
  -H "Authorization: Bearer pk_your_key" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Professional invoice with logo, line items, tax, payment terms"}'

AI generates a complete, styled HTML template you can reuse.

Fill a Template with Data Automatically

curl -X POST https://pdfgen-api.vercel.app/api/ai/fill \
  -H "Authorization: Bearer pk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "template": "<your-html-template>",
    "data": {
      "company": "Acme Corp",
      "items": [
        {"name": "Web Development", "amount": 50000},
        {"name": "Hosting", "amount": 12000}
      ],
      "tax_rate": 0.18
    }
  }'

AI maps your JSON data to the template fields — no manual field mapping needed.

JavaScript Example

const response = await fetch(
  "https://pdfgen-api.vercel.app/api/generate",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer pk_your_key",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      html: '<h1>Invoice #001</h1><p>Total: $1,500</p>',
    }),
  }
);

const { url } = await response.json();
console.log("PDF ready:", url);

Python Example

import requests

response = requests.post(
    "https://pdfgen-api.vercel.app/api/generate",
    headers={
        "Authorization": "Bearer pk_your_key",
        "Content-Type": "application/json",
    },
    json={
        "html": "<h1>Receipt</h1><p>Amount: $250</p><p>Status: Paid</p>"
    },
)

print("PDF ready:", response.json()["url"])

The Stack

Here's what powers PDFGen AI:

Layer	Technology	Why
Hosting	Vercel (serverless)	Zero config, auto-scaling
Framework	Next.js (App Router)	API routes + frontend in one
Auth + DB	Supabase	PostgreSQL, auth, file storage
PDF Rendering	Puppeteer + @sparticuz/chromium	HTML to PDF in serverless
AI	AWS Bedrock (Nova Micro)	Fast, cheap template generation
Billing	Lemon Squeezy	Merchant of Record

The Chromium Challenge

The hardest part was getting Puppeteer to work on Vercel serverless. The standard Chromium binary is too large. Here's what worked:

Use @sparticuz/chromium — stripped-down build for serverless
Add outputFileTracingIncludes in next.config.ts to bundle the binary
Launch with headless: "shell" mode for faster startup
Disable GPU with setGraphicsMode = false

Cold-start PDF generation: under 5 seconds.

Why Supabase?

Auth — email/password with magic links, zero config
Database — PostgreSQL for API keys, usage tracking
Storage — PDFs stored with signed URLs
Free tier — generous enough for an MVP

Why AWS Bedrock?

I originally used the Anthropic API directly, but switched to Bedrock:

Pay-per-use — no monthly minimums
Amazon Nova Micro — fast, cheap, perfect for templates
Bearer token auth — simple, no complex AWS SDK needed

5 Built-in Templates

PDFGen AI comes with ready-to-use templates:

Invoice — line items, tax, totals
Receipt — clean payment receipt
Report — business report with sections
Certificate — achievement/completion
Letter — formal business letter

Lessons Learned Building Solo

Start with the API, not the UI. I tested with curl for weeks before building the frontend. Forced me to get the DX right.

Free tiers are your friend. Total infra cost: ~$0/month. Vercel free, Supabase free, Bedrock pay-per-call.

Billing in India is tricky. Stripe doesn't support Indian merchants for international payments. Lemon Squeezy acts as Merchant of Record — handles global payments, pays you out.

SEO from day one. Sitemap, robots.txt, JSON-LD, OG images — all added before launch. 10x easier than retrofitting.

Ship fast. Idea to production in 4 weeks. Not perfect, but working.

Pricing

Plan	Price	PDFs/month	AI calls/month
Free	$0	50	10
Starter	$19/mo	2,000	100
Pro	$49/mo	15,000	500
Business	$99/mo	100,000	2,000
Enterprise	$299/mo	Unlimited	5,000

No credit card required for free tier.

What's Next

More built-in templates based on user requests
Webhook notifications when PDF is ready
Batch generation — hundreds of PDFs in one call
Custom font uploads
PDF merging

Try It

The API is live and free to start:

Website: pdfgen-api.vercel.app
Docs: pdfgen-api.vercel.app/docs

I'd love feedback — especially on the API design and developer experience. What would you build with it?

Built solo with Next.js, Supabase, AWS Bedrock, and too much coffee.

DEV Community