If you sell clothes in Egypt, your shop is an Instagram page and your checkout is a DM. That's how TRIPLE, a men's streetwear brand I work with, was running: every order was a chat thread, every size question a voice note, and half the "orders" never showed up to pay.
A few months ago we decided to fix that with a real store. The catch: it had to feel native in two languages at once. Egyptian shoppers browse and search in Arabic, the brand's lookbook culture is English, and neither audience should feel like an afterthought.
This is what we shipped and what I learned building it. The store is live at triplestore.vercel.app if you want to click around while reading.
What we shipped
A full e-commerce store for a streetwear catalog: ~50 products across categories like sweatshirts, tees and jeans, a cart, cash-on-delivery ordering with a deposit step, an order-tracking page for customers, printable receipts, and an admin dashboard the brand runs themselves - products, prices, coupons, stock, size charts, even the SEO copy, all editable without touching code.
And the part this post is about: every single page exists twice. English at /, Arabic at /ar. Not a translation plugin, not a language cookie that swaps strings - two real, linkable, indexable versions of the whole store.
The stack
- TanStack Start (React, SSR) - server-rendered pages, because a store that Google can't read doesn't exist, and most Egyptian buyers are on mid-range phones on mobile data.
- Supabase (Postgres + storage) - products, orders, inventory and images. Row-level security means the storefront reads through policies and the admin writes through policies; there's no separate API layer to secure twice.
- Vercel - hosting and previews.
Nothing exotic. The interesting decisions are in the bilingual and payments work, so let's go there.
Two versions of every page, not a translation toggle
The decision I'm happiest about: Arabic is not a string swap on the same URL. Every page exists at /... in English and /ar/... in Arabic, and both are first-class routes.
/shop -> English catalog
/ar/shop -> Arabic catalog (RTL)
/product/essential-tee
/ar/product/essential-tee
Three reasons. First, SEO: Google indexes Arabic search queries against Arabic pages. An Egyptian shopper types "هوديز رجالي" (men's hoodies), not "hoodies" - and a language cookie gives you nothing to index against that. Second, sharing: when the brand posts a product link in an Arabic caption, the link itself is Arabic. Third, honest rendering: <html dir="rtl" lang="ar"> on the Arabic routes, which makes the browser do the right thing with text direction, punctuation and form fields instead of you fighting it in CSS.
The sitemap carries both versions with proper alternates, 63 pages in each language:
<url>
<loc>https://triplestore.vercel.app/shop</loc>
<xhtml:link rel="alternate" hreflang="en"
href="https://triplestore.vercel.app/shop"/>
<xhtml:link rel="alternate" hreflang="ar"
href="https://triplestore.vercel.app/ar/shop"/>
<xhtml:link rel="alternate" hreflang="x-default"
href="https://triplestore.vercel.app/shop"/>
</url>
Each page also cross-links its twin with <link rel="alternate" hreflang> in the head, and canonical points at itself, never at the other language. The two versions are siblings, not copies.
RTL is a design task, not a CSS property
Setting dir="rtl" gets you 80% of the way. The remaining 20% is where bilingual sites usually look broken, and it's all manual:
-
Use logical properties everywhere.
margin-inline-startinstead ofmargin-left,inset-inline-endinstead ofright. If the codebase does this from day one, most of the layout mirrors itself for free. -
Directional icons must flip, semantic icons must not. A back arrow points right in RTL. A WhatsApp glyph or a size chart does not. We ended up with a small
FlipInRtlwrapper component so the decision is explicit at every call site. -
Numbers stay Western. Egyptian shoppers read prices as
650 ج.م- Western digits, Arabic currency label, Arabic amount formatting around it. Full Arabic-Indic digits (٦٥٠) would look more "translated" and less familiar. - Fonts carry the brand in both scripts. The store uses Anton for Latin display type and Alexandria for Arabic copy. Arabic display fonts are a much smaller pool than Latin ones; pick early, because the whole visual identity shifts with that choice.
- Test the forms, not just the hero. A checkout form mixing an Arabic name, an Egyptian phone number and an English email address is where direction bugs actually hurt - placeholder alignment, input direction, validation messages.
The Arabic home page, fully RTL with mirrored navigation:
Payments: build for how Egypt actually pays
Card penetration is low and cash on delivery is the default expectation, but pure COD has a brutal failure mode: fake orders. Someone orders three hoodies to a made-up address, the courier rides out, nobody's home, and the brand eats the shipping both ways.
So COD orders require a small deposit - 100 EGP by Vodafone Cash, the mobile wallet basically every Egyptian phone has. The checkout generates the order in an awaiting deposit confirmation state, the buyer transfers 100 EGP and uploads a screenshot of the transfer, and the order only enters the fulfillment queue when the brand confirms the deposit from the dashboard. The deposit comes off the final price on delivery.
A product page in Arabic - price in EGP, sizes, quantity, stock state:
Is it friction? Yes, deliberately. The bet is that a buyer who won't pay 100 EGP (~$2) up front was never picking up the package. The brand's answer so far: the fake-order problem simply stopped.
Everything else about the order flow is built for the same reality: customers can check their order status on a public "my orders" page without an account (orders tie to phone numbers, not logins), receipts print cleanly for the courier bag, and the whole thing works on a mid-range Android over mobile data.
The unglamorous parts that mattered
- SEO is generated, not written. Product titles, descriptions, structured data and sitemap entries are generated from the product data, with the live price in the JSON-LD. When the brand edits a price in the dashboard, the SEO layer updates with it - nobody maintains 126 pages of meta by hand.
- Inventory is strict. Stock decrements per size, sold-out items stay visible and indexable marked as sold out rather than 404ing, and the admin warns (without blocking) when a product is missing a description, Arabic name, category or stock count.
- The admin is the product. The people running this store are not developers, so size charts, shipping fees per governorate, coupons and even the brand copy are dashboard settings. Anything hard-coded is a support message waiting to happen.
What I'd do differently
- Start RTL on day zero. We built English-first and retrofitted. Logical properties and the icon wrapper would have been nearly free early; they were a week of diff later.
- Translate the microcopy with the brand, not for it. Egyptian Arabic has its own shopping vocabulary, and a literal translation of "checkout" or "sold out" reads wrong. The final pass was the brand's owner reading every string out loud.
- Deposits are a product decision wearing a payments costume. It changed who orders, not just how they pay.
The store is live at triplestore.vercel.app (try /ar for the Arabic version). Happy to answer questions about the bilingual setup, the deposit flow, or TanStack Start + Supabase for e-commerce in the comments.



Top comments (0)