I built a WhatsApp-to-book pipeline with Make.com, OpenAI and Lulu.com — here's the full stack

Marvin Munos — Wed, 08 Jul 2026 14:06:25 +0000

Minha Historia turns your parents' voice notes into a hardcover memoir book. The French version is Mon Livre de Vie. Here's how the entire pipeline works, with no custom backend.

The concept

We send one question per week via WhatsApp
Parents answer with a voice note (60 seconds average)
We transcribe, edit, and compile 36 answers into a hardcover book
Every chapter has a QR code that plays the original voice recording

The stack (zero custom backend)

Shopify (Dawn theme) — storefront, payments, order management
WhatsApp Business Cloud API — question delivery and voice note collection
Make.com — 10 scenarios orchestrating the entire pipeline (question scheduling, response collection, transcription triggers, relance sequences, PDF generation)
OpenAI API — transcription (Whisper) + editing (GPT-4o for grammar and flow without changing the person's voice)
Airtable — client database, response tracking, progress dashboard
Brevo — transactional emails and marketing sequences
Cloudflare R2 — audio file storage (cheap, fast, S3-compatible)
Cloudflare Workers — PDF interior generation (custom worker iterating through chapters)
Lulu.com — print-on-demand hardcover books, ships worldwide

Why WhatsApp and not an app

The target user is a 70-80 year old parent in a small Portuguese village. They don't install apps. They don't create accounts. But they send WhatsApp voice notes to their grandchildren every single day.

WhatsApp is the only interface. No app, no login, no dashboard for the parent. They just receive a question and talk.

The Make.com architecture

10 scenarios handle everything:

S1: Order webhook from Shopify → create client in Airtable
S2: Weekly question scheduler → send via WhatsApp Business API
S3: Response receiver → transcribe via Whisper → store in Airtable
S4-S5: Relance sequences (gentle reminders if no response after 48h)
S6: Progress tracking → dashboard updates
S7: Transcription editing → GPT-4o polishes grammar without changing voice
S8: PDF generation trigger → Cloudflare Worker builds the interior
S9: Print order → Lulu.com API
S10: QR code generation → links each chapter to its audio file on R2

Cloning to a second market

The French version (Mon Livre de Vie) took 2 weeks to clone because the stack is market-agnostic. The only real work was translating the 36 questions, the email sequences, and the Shopify theme. The WhatsApp flows, Make scenarios, and printing pipeline didn't change at all.

The product

Portuguese version: minhahistoria.pt — 99€ one-time, no subscription
French version: monlivredevie.fr — 114€ one-time, no subscription
No app to install, works on any phone with WhatsApp
4 hardcover designs (Terra, Mar, Serra, Vinho), gold ornaments, premium typography

Also building SiftedTools, an independent AI tool comparator for e-commerce in 7 languages.

Would love feedback from anyone who's built WhatsApp-based products or print-on-demand pipelines.

How I built a 145-page multilingual site with Astro + Cloudflare Pages (and why I didn't use Next.js)

Marvin Munos — Wed, 08 Jul 2026 14:01:58 +0000

I recently launched SiftedTools, an independent AI tool comparator for e-commerce, published in 7 languages (FR/EN/ES/DE/IT/PT/NL). Here's the technical stack and why I made these choices.

The stack

Astro for static site generation (145 pages, builds in under 2 seconds)
Cloudflare Pages for hosting (free tier, drag-and-drop deployment)
Markdown for all content (65 money pages + 21 guides + 7 homepages)
GitHub for version control
No database — everything is flat files

Why Astro over Next.js

For a content-heavy site with zero interactivity (no user accounts, no forms beyond a newsletter signup), Astro ships zero JavaScript by default. That means:

Perfect Lighthouse scores out of the box
Pages that load in under 500ms on any connection
No hydration cost, no client-side routing overhead

Next.js would have shipped 80-200KB of JS for features I don't use. For a comparison site where SEO is everything, that's unacceptable overhead.

The i18n architecture

7 languages live under subfolders (/fr/, /en/, /es/, etc.) with:

Localized slugs (not just translated content, but native URL structures)
Full hreflang network (8 entries per page: 7 languages + x-default)
A neutral root language selector at / with soft browser detection
Each content page exists as clean .md by appending .md to the slug — this is specifically for LLM crawlers who prefer Markdown

The i18n dictionary lives in a single TypeScript file with ~600 keys across 7 languages.

The GEO layer (AI search optimization)

Beyond traditional SEO, I optimized for AI search engines (ChatGPT, Gemini, Perplexity):

robots.txt explicitly allows GPTBot, ClaudeBot, PerplexityBot, and 7 other AI crawlers
llms.txt at the root indexes all 145 pages with descriptions (the emerging standard for LLM discoverability)
llms-full.txt contains detailed verdicts for deeper context
Every content page as .md endpoint — LLMs parse Markdown better than HTML
Schema.org with FAQPage, author, Organization on every money page

Deployment

Cloudflare Pages free tier with drag-and-drop. No CI/CD, no build pipeline. I run npm run build locally, drag the dist/ folder to the Cloudflare dashboard, and it's live in 30 seconds. Simple, free, fast.

The entire site costs $0/month to host.

What I'd do differently

I'd set up CI/CD from day one (GitHub Actions → Cloudflare Pages) instead of manual drag-and-drop
I'd use Astro content collections with type-safe schemas earlier (I retrofitted them)
I'd build the 7-language architecture before writing content, not after

Try it

The site: https://siftedtools.com
The repo: https://github.com/MVMStudio34/siftedtools

Happy to answer questions about the Astro i18n setup or the GEO optimization layer.

DEV Community: Marvin Munos