All That AI
Since 2026.09.01 338 guests aboard 32 repeat customers 4,120 site visitors 8,736 page views 114 waitlist entries · 168 people
ALL THAT AI / AI BOOKING & OPERATIONS

This AI booking system
you are looking at — we built it

Geobuk Water Sports runs on an AI booking system we designed, built and still operate. Live seat counts, waitlists with seat holds, automatic bank-transfer matching, bundled notifications and refund math — one piece, not eight plugins. Instead of a slide deck, here is a thing that runs: this site.

🎣 207 bookings handled🔔 114 waitlist entries · 168 people💳 Deposits matched automatically🤖 AI assistant built in🌐 Multilingual + SEO
Get in touch See allthatai.kr

No demo needed — just look around

Everything below is live on this site, taking real bookings from real customers right now.

Live seat countsRemaining seats per departure, in the calendar. Overbooking is impossible.
Waitlist with seat holdsWhen a trip is full, customers join a queue. If someone cancels, the owner is told who is next — and that seat is locked while the call happens, so nobody else takes it.
Automatic deposit matchingBank SMS is parsed and attached to the right booking. The screen shows in colour whether a payment was matched automatically or by hand.
Payment deadlinesUnpaid bookings raise a reminder card on their own, with a payment link and a ready-to-send message.
Refund math, up frontThe customer sees what cancelling costs them before they cancel. Fewer arguments on the phone.
One notification, not nineNine guests used to mean nine pings. Now it is one card with checkboxes.
AI assistantAnswers customer questions from the actual programme, tide and policy data — grounded, so it does not invent.
One admin screenManifest, payments, revenue, catch reports and server health in one place.
Search and languagesPer-programme landing pages plus English, Japanese and Chinese, generated automatically.

This site, by the numbers

338guests Actually stepped aboard — 39 departures

Average 8.7 per departure capacity 10

People who actually boarded since 2026-09-01 - headcount, not bookings.
32repeat Booked twice or more 22% Out of 144 customers · most frequent: 7 trips
168people114 entries Queued because there was no seat 10.9 2210.10 810.11 210.17 210.18 310.24 1 10 seats Each departure has 10 seats (dashed line). A bar past the line means more people are waiting than that day can carry. Entries and people differ — a party of two needs two seats.
A long waitlist means you are out of seats to sell. A one-boat operator still has 114 parties — 168 people — queued. Run that queue on paper and the order gets lost and cancelled seats sail empty. Let the system hold the seat and remember the order, and that seat turns into revenue.

Who this fits

  • Fixed-capacity, scheduled departures — charter boats, yachts, diving, surfing, riding, workshops, studio rentals
  • Deposits by bank transfer — no card fees, but confirmation still automatic
  • Frequent cancellations or no-shows — showing the penalty before they cancel reduces disputes
  • Owner-operators on site — the manifest and the money on one phone
  • Places that want foreign guests — English, Japanese and Chinese pages with search visibility
And where it does not fit: walk-in businesses with no reservations, or operators already happy with their platform.

Booking is not all we do

If a job has AI in it, the field does not matter — vision, language (LLM), speech, multimodal, physical AI and healthcare. Where it matters, we train a custom model and install it on an air-gapped network.

VisionJudging from images and video — automatic face masking, defect and grade classification, counting people or vehicles, PPE detection, species and crop sorting. — Catch photos on this site have faces found and covered automatically at upload. Nobody paints them by hand.
Language (LLM)Chatbots grounded in your own data, contract and document review, drafting, pulling line items out of email and SMS. — The chat button on this site, and the bank-SMS matching, are both this.
SpeechCall transcription, call-centre summaries and scoring, voice guidance (TTS), hands-free input on site, voice-phishing detection from call audio.
MultimodalImages and text judged together — site photo plus work log to flag defects, product photo plus copy to build a catalogue, video plus captions for scene search.
Physical AISystems that see and then move — driving perception, robot arms and mobile robots, machine anomaly detection, models pushed onto edge devices so they run on site. — Author of "Vision-Based Autonomous Driving and Physical AI, the Tesla Way" (840 pages).
HealthcareReading assistance, intake and charting automation, fall and abnormal-behaviour detection for elderly care, medication tracking. — Diagnosis itself is regulated as a medical device. We design these as assistive tools and stay on the right side of that line.
Custom modelsFine-tuned on your own data and, where required, installed on a network with no internet at all — for places where the data must not leave the building.
Automation and monitoringDaily scraping of tenders, grants or competitor pricing with eligibility filtering; services that notice they died, restart and report. — This booking server restarts itself, and says on screen whether the last shutdown was clean.
Training and advicePractical AI training for staff, and advice on what to automate first. — Lead instructor on a national KDT AI programme; teaches at universities and companies.
Not sure what to hand over? Name three things your people repeat by hand. We will tell you which of them AI can take, and which are cheaper left to a person. If it will not work, we say so.

How it gets built

A. As is

Similar business? Change the name, programmes and prices. Fastest and cheapest.

B. Adapted

Departure structure, pricing rules, deposit and refund policy, notification flow reshaped to your business. Most projects land here.

C. From scratch

Multiple sites, multiple vessels, inventory integration — designed from the ground up.

TimelineA: 1-2 weeks · B: 3-6 weeks · C: to be discussed
CostDepends on scope. One call is usually enough for a ballpark.
Running costThe server runs on a single PC if you want it to. Keeping fixed monthly cost near zero is the default.
OwnershipWhat we build is yours. Source and data are handed over.

Who you would be working with

NameLEE SEOK CHANG — Founder · Adjunct Professor / Specialist Instructor, KOREATECH
CompanyAll That AI (올댓에이아이)
WorkAI booking and operations systems, workflow automation, AI teaching
Address239 Wangsan-ro, 7F #731, Dongdaemun-gu, Seoul, Korea
Siteallthatai.kr

Talk to us

Tell us what you sell and how you take bookings today. A 20-minute call is usually enough to say whether this fits.

Not only software — if you run boats and want to talk about working together (shared bookings, overflow guests, joint operation), that is welcome too.

This page lives inside the Geobuk Water Sports site on purpose — so you can see the real customer-facing screens: the English landing page · the booking screen

✉️Email us