A guide the host makes once. A manual every guest reads in their own language.

01 — The problem

Unfamiliar spaces, unfamiliar devices, unfamiliar language

You arrive at a rental in France. You want to run the washing machine, turn on the heater, or find the right TV input — but every label is in French, and even in your own language the controls aren't obvious. The manufacturer's manual is gone, generic, or online in one language. So you guess, you break something, or you message the host at 11pm.

Hosts feel the same problem from the other side: the same questions over and over, appliances misused, and reviews that ding them for confusion. Their current fix — a paper binder or a wall of sticky notes — doesn't scale across languages and goes stale the moment they swap a kettle.

The gap isn't “there's no manual.” It's that no manual is specific to this unit, simple, and readable in the guest's language at the moment they're stuck.

02 — Who it's for

Two people, one product

This is two-sided, and being clear about who is the hero of creation versus the hero of the moment keeps the design honest.

The host — creator, and the one who pays

Busy, a little lazy (a paper binder is free), and motivated by concrete outcomes: fewer late-night questions, fewer broken appliances, better reviews, a more professional impression. The manual is a means to those ends, not the goal. So the whole creation flow has to be near-zero effort or the host never comes back.

The guest — where the “wow” happens

Wants to understand one specific device right now, in their language, without installing anything or signing in. The magic is: scan a code, pick a language, understand. That instant is what gets remembered and talked about — and it's what quietly recruits the guest's own host-friends.

03 — How it works

From a 10-minute walkthrough to a live multilingual page

The flagship capture flow is a voice walkthrough: the host walks the apartment talking — “this is the washing machine, always use 30°, don't overfill it or it leaks” — while snapping photos. That captures exactly the specific, tacit knowledge that is the entire value, with far less effort than typing.

The core artifact is a living, multilingual web page — not a PDF. Only a web page can hold a language switcher, short how-to videos, per-section scan analytics, and updates when the host swaps a device. A downloadable PDF stays as a secondary export for hosts who want something printed on the wall.

04 — Decisions locked

The spine we agreed on

05 — The first open question

Web or mobile app?

The trap here is treating this as one question. It's really two, because your two users have opposite needs — and once you split them, the answer is clear.

The guest side settles it

A guest will never install an app to read a washing-machine manual. The whole promise is “scan a code and understand in ten seconds.” That means the guest experience must be a plain web page opened straight from the QR scan. This alone rules out a mobile-app-only product.

The host side could go either way — but shouldn't, yet

Creation needs a camera and a microphone. Both work in mobile browsers today. A native app buys marginally better capture and offline recording, but at real cost: two codebases (iOS + Android), app-store review delays, install friction for hosts too, and slower iteration while you're still learning what the product even is.

06 — The second open question

Accounts & membership

You're right that you need login — but the sharp move is to notice only one of your two users signs in.

For host auth, keep it standard and boring: email + password plus a Google sign-in, ideally through a managed auth provider so you're not hand-rolling security. That's plenty for v1.

The tiers — and where to put the paywall

One caution on your instinct to put saving and editing behind premium: if the free tier can't save or edit, then free isn't a working product — it's a dead demo. That strangles the two things freemium is supposed to buy you: guests scanning live QR codes are your best free advertising, and a host won't pay until they've felt the magic with a real guest. Gate on scale and polish, not on whether the core works.

07 — Content model

Categories as scaffolds, specifics as substance

Templates organized by category — accommodation (house, hotel room), machines, tools — are useful, but as scaffolds that tell the AI what to ask, not as the content itself. A generic “load clothes, add detergent” step is worthless; the guest already knows that. The value is “this machine leaks if overfilled; use the third button.”

A workable shape for the data:

• Property — the space, its default language, its QR code, its published guide.
• Section — one device or topic (washing machine, heater, check-out, trash).
• Step / note — text plus an optional photo or short video.
• Translations — auto-generated per-language versions of the approved source.

08 — Trust & safety

The one human check, placed where it counts

Two different things could be “wrong”: the translation could drift from the source, or the source itself could be wrong for this device. Modern models handle the first well. The second is the dangerous one — and only the host can catch it, because only they know the machine. So the host reviews and approves the source, once, in their own language (fast — it's their apartment), and translation flows automatically from there.

09 — Architecture sketch

Rough shape of the build

High level only — enough to reason about cost and scope, not a final spec.

• Frontend — one responsive web app / PWA (e.g. a React-based framework). Serves both the host studio and the public guest pages.
• Guest pages — fast, cached, served from a CDN, since guests are worldwide and load-time is the experience.
• Backend / API — handles accounts, properties, guides, membership and billing state.
• AI pipeline — speech-to-text for the voice walkthrough; a language model to structure and draft; translation for target languages; optionally vision to read control-panel photos.
• Object storage — photos, videos, audio, and generated PDFs (your Google Cloud Storage buckets fit here).
• Database — relational (e.g. Postgres) for hosts, properties, sections, translations, memberships.
• QR + payments — a unique URL + QR per guide; a subscription billing provider (e.g. Stripe) for premium.

10 — Scope

The v1 MVP — prove one thing

The single thing v1 must prove: a host can capture a guide with almost no effort, and a guest scans it and genuinely understands, in their language. Everything not serving that proof waits.

11 — Open questions

Assumptions worth testing before betting big

Will hosts actually record a voice walkthrough? Or will they feel silly and default to typing? Test with a handful of real hosts before building the polished flow.

Is AI translation good enough for appliance / technical wording across your target languages? Check real instructions, not marketing copy.

How specific can vision get from a photo of a control panel? This decides how much the host must narrate vs. how much AI reads for them.

Which languages first? Driven by each host's actual guest mix, not a global guess.

What's the right paywall line and price? How many properties / languages are free before value tips into paying?

How do the first hosts find you? Distribution is unsolved — host communities, Airbnb/Booking host groups, word of mouth from live QR codes.

12 — Next steps

Where to point energy first

• Validate the capture flow with 3–5 real hosts — even manually, “concierge” style, before writing much code.
• Stress-test translation quality on genuine appliance instructions in your top languages.
• Build the MVP above as one responsive web app, host accounts only, guest pages public.
• Decide the free/premium line from what you learn, then wire up billing.