SD Realty is a project of our own, part of the Sebastian Desigual family of brands: an agency in Belek, Antalya, offering long-term and short-stay rentals, services for tenants, and support in three languages. With no external client, we could set ourselves a high bar — not a prototype, but an agency ready to launch. The original plan had the final stages at the end of November. We finished on 14 October, more than six weeks early.
What we ended up with
- Brand: positioning, three pillars — Property · Lifestyle · Investment — a logo in 5 versions and 5 colours, a 20-page brand book, and a design system with two themes and 14 components.
- Website: 17 pages in Russian, English and Turkish — 51 addresses in the sitemap, enquiry and booking forms, multilingual SEO. Lighthouse Accessibility: 100.
- 5 automations: website enquiries to a sheet and Telegram, an AI assistant, a services bot with booking, a Mini App, and property descriptions on n8n.
- Collateral: 13 types, about 50 print-ready layouts — from foil business cards to a signage sketch — plus a QR system with 15 short links.
The order of work
Days 1–2. Plan, name, positioning, site structure, references. The busiest day of the sprint was the second: the whole brand and version 1 of the site layouts.
Days 3–5. Version 1 of the site in code, the first enquiry automation, the brand book. We built enquiries in Make first, then two days later did the same in code. In our test, code responded to the site in 628 ms and Make in 950 ms. That's one measurement, not statistics, but it backed up the conclusion that code is faster and has no operation limits; code is what runs in production.
Days 5–7. The AI assistant, district pages, version 2 of the site, and a review of the site by a separate agent.
Days 7–8. The services bot and the Mini App — one session each, because the spec was ready in advance. In live testing the Mini App passed 10 out of 10 scenarios.
Days 8–13. Property descriptions on our own n8n, collateral, QR codes, mock-ups, the overview and the agency deck.
The pipeline behind the speed
The speed came from the way we worked, not from rushing. The same order applies to every stage:
- Brief → spec → design → code or no-code → checks → log. Every coding stage started with a spec that already recorded the decisions and a list of checks — 13 each for the assistant and the services bot.
- First 2–3 samples, then the rest after sign-off. Revisions stay cheap and nothing gets redone from scratch. Over the whole sprint, nothing was.
- Moving between tools goes through a file. Chat, Design, Code, Make and n8n can't see each other's history, so briefs, specs and the results of each tool are written to project files. Over 110 steps, no tool lost context.
- A separate agent checks finished work. It catches what the author misses: the site was open to search engines too early, the language markup pointed the wrong way, a long word didn't fit a button on mobile.
What went wrong
Mistakes became part of the pipeline too — each one turned into a rule.
- Two coding sessions in one repository at once tangle the branches. Now it's one at a time.
- A cheap AI reseller was 6 times cheaper than the official API, but response times went from seconds to a minute. Client bots use the official API only.
- Hosting quietly stopped building the site after changes. Now we check the build has started after every push.
- An internal note meant for the AI ended up in a description. Such notes are now marked "don't mention in the text".
What carries over to your project
Almost all of it. A website foundation with languages, forms, SEO and cookie-free analytics. Two ways to automate enquiries — a Make workflow you can edit yourself, or a code module. An assistant knowledge base and rules, with measured running costs. A booking bot with four modes, suited to salons, schools and service businesses. Templates for proposals, letterheads and decks.
A typical order doesn't take 13 days: a short brief, a template, two rounds of revisions. SD Realty is an "agency end to end" project, and there's no need to measure an ordinary landing page against it. But the pipeline is the same.