Deterministic Trip Planning: Put the Catalog First

The obvious version of Ember Coast is a thin wrapper around a language model. You type "romantic week in Hawaii," it writes you a dreamy day-by-day, you paste in a credit card. I deliberately did not build that.

Here is why, and what I built instead.

Ember Coast began as Aurum Atlas, a hand-planned travel service built around a spreadsheet. I moved the repeatable selection and scheduling work into software so the product would not depend on me assembling every itinerary. Human-led concierge work stayed a separate service.

The rename came with the reframe. Ember Coast is not a travel agency with a website. It is a planning engine with a concierge desk attached for the people who want one.

I kept itinerary selection out of the language model because a plausible plan is not enough. Dates, travel time, and operating schedules need explicit constraints. A model can assist with research or wording, but fluency should not decide whether an activity belongs on Tuesday.

Here is the whole thesis. The plan is assembled by code, not written by a language model. There is a file called buildPlan.ts that takes your dates, your travelers, your pace, and the moments that matter to you, and composes a dated itinerary out of a vetted catalog. It is boring, deterministic software. Given the same inputs and catalog version, it produces the same plan.

I know how that sounds in 2026, when everyone is gluing an LLM to everything.

A seasonal example needs actual dates, not an assumption about ski towns. Aspen's Silver Queen Gondola does run in summer, as the operator's published schedule makes clear. The useful question is whether the specific lift operates on the traveler's date and within the plan's timing.

None of that is the model being careless. A real itinerary has to survive contact with reality: operating seasons, opening hours, travel time between two points, whether a place still exists. That is bookkeeping, not creativity. So I made it bookkeeping.

The engine selects catalog entries rather than inventing them, and applies the operating information recorded for each entry. That makes the selection explainable. It does not make the data permanently true: maintenance, weather, sold-out slots, and changed hours still require a current operator check.

This is not an allergy to models. Codelit's prompt-to-diagram pipeline hands an LLM a sentence and gets back an architecture diagram, and that works because the output is a draft you are expected to edit. A booked Tuesday in Aspen is not a draft. Stakes decide the architecture, which is the same reasoning behind DateSafe, where provider latency must stay off the duress screen.

There are three ways to get a luxury itinerary: a chatbot writes it, a human plans it by hand, or an engine assembles it from a vetted catalog. Ember Coast is the third, and the price it pays for that is coverage.

That last cell is a real limitation and I would rather say it out loud. Catalogs grow slowly because vetting is slow, so Ember Coast will tell you it does not cover a place instead of improvising a week there.

Ember Coast takes four inputs: your dates, who is coming, the pace you want, and the moments that matter most. It hands back a dated plan with a vetted stay, table, and experience for every day of the trip.

• A dated, day-by-day itinerary built around your people and your pace, not a generic template.
• A vetted stay, table, and experience for every single day, each one something I would send my own family to.
• Official direct booking links, at list price, with no markup.
• A realistic cost estimate, a printable PDF, and a copy in your inbox.

The pacing logic is the part I fuss over most. A good trip breathes. The engine spaces out the big days, protects a slow morning after a late night, and refuses to stack three long drives back to back. Those are the rules a person who plans a lot of trips follows without thinking, written down and enforced, so every plan follows them even when I am asleep.

The human work in Ember Coast sits upstream of the engine. I decide what enters the catalog and I write the pacing rules the engine enforces, then the software owns the scheduling, because scheduling is the part with a right answer.

That split keeps showing up. ViralVault publishes a tech briefing twice a day on rails I set, and the automation is the easy half: editorial judgment stays mine. AbleMakers holds a door open for makers with Down syndrome by handing software the mechanical work, so the person can do the part only they can do.

Ember Coast charges a flat planning fee and provides official booking links without adding markup to those bookings. Separating the planning fee from the operator keeps the commercial model understandable. Travelers still need to check the final price and terms on the booking page.

I do not mark up your travel. What you book is what it costs.

If you want the logistics handled, there are concierge tiers on top. Signature Escape adds premium stays, flight coordination, and transfers. Private Atelier adds private drivers, guides, and chefs. Estate Reserve is for staffed villas and multi-household trips. Those are real work and they cost real money.

But they sit beside the plan, never inside it. A planning estimate and a confirmed booking price remain different numbers.

Ember Coast covers seventeen destinations across seven western states, from the Hawaiian islands to Aspen to the Napa and Sonoma valleys, with new ones vetted before they ship rather than after. The founding plan is twenty-nine dollars and the concierge tiers go up from there.

The tagline is "your private trip, planned to the day." The engineering version of that promise is less romantic and more important. Every recommendation comes from a defined catalog and an explicit scheduling rule. The remaining work is keeping that catalog current and making the difference between a recommendation and a confirmed reservation obvious.

If you are building anything in this shape, the useful lesson is which half is hard. The catalog is the asset. The model is the part you bolt on later, if it turns out you need one at all.

If you are planning something that matters, start a plan. Start with your dates and preferences, inspect the proposed days, and confirm the details with the operators before booking.