
Why I Created Korail
Why I created Korail, with built‑in vector search à la Perplexity.
Why I Created Korail
Why I created Korail, with built‑in vector search à la Perplexity.
You’re promised “AI‑created websites” in one click. Page, image, and text generators for everything. But in the end… it’s just an empty shell nicely served.
I wanted something else.
Not a site generated by AI. A site that is AI.
A site that understands. That answers. That evolves. And that doesn’t waste your time.
I wanted to be able to:
- launch a site in minutes,
- make it visually impactful without useless animations,
- have it speak multiple languages without messy plugins,
- and have it answer from day one.
No triple backflips in parallax. No frills to hide emptiness. Just a clear, fast, intelligent site.
And above all: useful.
Today, GPT can answer (almost) anything. So why can’t your own site do the same? Why force people to navigate 12 menus to find an answer your site could give in one clear sentence?
I wanted a system that does all that, fast. Couldn’t find it. So I built it.
An embedded vector engine, Perplexity‑style
In Korail, every piece of content is automatically indexed for vector search, thanks to two combined models:
text-embedding-ada-002
(1536 dimensions) — for speedtext-embedding-3-large
(3072 dimensions) — for deeper semantics
Every sentence and paragraph is chunked and transformed into a sequence of numbers (vectors), and stored in Supabase with pgvector — a database designed for this kind of search.
Result? The site doesn’t look up keywords. It understands the meaning of a question. It retrieves relevant passages even when words differ, even with typos, even when phrasing is fuzzy.
That’s what lets Korail work like a personal Perplexity — based on what the site already contains.
Even with a single page, a site can answer. And the more content you add, the more precise it becomes.
And it works in every language
Someone can ask a question in Portuguese… Korail will answer in Portuguese — even if your content is in French or English.
Vector search is natively multilingual: it detects intent, not language. I live and work in French, English, Spanish (tourism in Bacalar, Mexico). So it was non‑negotiable: multilingual had to be integrated, fast, and frictionless.
Enriched answers, Perplexity‑style
Answers aren’t just plain text blocks. Korail can return structured, verifiable answers:
- Clear text (synthesized from vector search),
- Clickable sources to the relevant pages,
- Images from your media library,
- Videos embedded,
- Audio/podcasts embedded,
- and, if needed, other media or files linked to the answer.
Important: formatting uses GPT‑4 mini (or GPT‑4) to compose the answer, exclusively from passages found by the internal vector database. No external invention. No searching elsewhere unless you decide to.
Built‑in AI, where it truly helps
In Korail, AI isn’t a gimmick. It’s integrated where it actually saves time.
1. Answer clearly
When vector search surfaces the right passages, GPT‑4 mini (or GPT‑4) crafts a clear, readable, direct answer. But crucially: it never goes elsewhere. It only reformulates what your site already contains. No hallucinations. No inventions.
2. Help you create faster
AI can also assist with content creation:
- propose a catchy title,
- suggest a page outline,
- generate a clear summary of an article,
- translate automatically into 12 languages (without breaking layout).
Always in context. Always editable.
AI in Korail doesn’t replace what you write. It amplifies your work, and saves you from repetitive tasks.
A clear, multilingual, frictionless interface
With Korail:
- you can edit directly on your site,
- or use a full online editor when you want to manage everything at once,
- you publish in one click,
- you translate natively into 12 languages,
- and you enable smart search that works in each of them.
You can start with a single page and already‑active vector search. Then progressively add a blog, forms, a bookings module (in development, targeted for September), or even a store (to follow).
A solid, modern, familiar stack
Korail isn’t built on some exotic tech. It’s a modern, proven stack used by leading web platforms.
- React — the library behind Netflix, Instagram, Airbnb
- Next.js — used by TikTok, Twitch, Hulu for performance and hybrid rendering
- Supabase — database, auth, storage; a modern Firebase alternative
- pgvector — vector search directly in PostgreSQL
- TailwindCSS — the design framework used by GitHub, Shopify, Vercel
- Stripe — reliable, secure payments
👉 Result: a robust system, easy to maintain, that evolves without breaking.
It’s my favorite stack — I can build fast, deploy without friction, and add new bricks as needs arise.
Easy migration from WordPress, Wix, Squarespace
I’ve often had to extract sites from WordPress, Wix, or Squarespace. So Korail ships with an importer built for speed and cleanliness.
- Pages & posts: content, slugs, dates, authors
- URL structure: preserve/rebuild paths, with optional 301 redirects
- Media: bring images over and rewrite links (hosted on your side)
- Taxonomies: categories, tags, menus — mapped to Korail blocks
- Basic SEO: titles, descriptions, canonical tags respected
- Internal links: auto‑repair where possible; warnings when broken
- Multilingual: detect versions (if present) and prep for translation into 12 languages
At the end, you get a block‑editable site with vector search already active. I keep refining it, but it’s already very useful for most redesigns.
Want to start from scratch? Possible. Want to migrate and improve gradually? That’s exactly the path I designed for.
And the name: Korail?
My personal site is Set Code & Sail. Because I sail between the turquoise waters of Bacalar and my lines of code… I wanted to keep that marine touch.
Coral is more than a backdrop: it’s a living, essential structure that shelters a quarter of marine biodiversity, protects coasts, feeds millions… and is disappearing before our eyes.
So yes, Korail is a wink. But it’s also a direction: build systems that are alive, useful, durable.
And if tomorrow this project can connect to a real reef‑preservation initiative, that’d be the perfect loop: helping protect a fragile ecosystem while building digital ecosystems.
Related posts
From Wix to Korail: how I import 500+ pages cleanly — Part 2
Migrating hindouisme.org: custom Node.js scraper, Korail importer, JSON → React blocks.
When 5,000 Years of History Meet New Technologies — Part 1
Rebuilding hindouisme.org and creating Korail: a vector-native CMS with React, Supabase, SEO, performance, simplicity.
Vector Playground – When AI shows how it understands
A visual, playful exploration of cosine similarity and semantic vectors — to see how AI connects ideas.
Marie Fa
Full-stack developer from Québec, based in Bacalar — sailing between turquoise water and lines of code.
Founder of Murmure & Korail, SocialRally and Adorable Sailing.