AI · Travel

I designed an AI travel agent that plans your entire trip — down to which api books the flight

Trip planning is broken. Wander doesn't help you research a trip — it prepares the whole journey for you.

Wander AI trip planner showing a complete Lisbon itinerary

trip planning is broken. You open 23 tabs, compare flights that expire while you're looking at hotels, guess whether two attractions are actually near each other, and after six hours you have a spreadsheet — not a trip. We accepted this as normal. It's not normal, it's only old. So I designed Wander: an AI travel planner that prepares the whole journey for you.

What Wander does

You tell it one sentence: "four slow, food-focused days in Lisbon, boutique hotels, no rushed mornings." That's it. Wander turns that into structured preferences — budget, pace, food style, accessibility, even what time you like to start your day — and then builds the entire trip: live flight options, hotels that match your vibe and budget, airport transfers, restaurant reservations, attractions, optimized daily routes, weather-aware scheduling, a full trip budget, and calendar events waiting in your phone.

The key word is coherent. This isn't a chatbot handing you a list of "top 10 things to do in Lisbon." It's a digital travel designer that makes every piece work with every other piece.

The part that makes it smart

Here's my favorite design decision: the AI never invents anything. No made-up flight times, no hallucinated opening hours. It pulls verified data from real services, and its job is to evaluate, not imagine.

Flights and hotels come from the amadeus travel apis — the same search-to-booking infrastructure the travel industry runs on. But Wander doesn't only rank flights by price like every booking site. It scores total journey time, connections, arrival time, baggage, cancellation terms, and even distance from your hotel. Because the $40 cheaper flight that lands at 1am and costs you your first morning? That's not cheaper.

Google places and Google routes handle the geography. The route engine compares travel times between every stop, so Wander notices that three places you'd love are in the same neighborhood and schedules them together — instead of bouncing you across the city like most itineraries do.

And the weather layer is the detail I'm proudest of. Wander checks hourly forecasts and plans around them: garden on the sunny morning, museum during the rain. If the forecast changes mid-trip, it reorganizes your day and tells you: "rain is expected monday afternoon, so I moved maat to 2:00pm and placed your river walk in the clearer morning window." That's not a feature, that's having a person who cares.

Same logic in the budget engine. Go over budget and Wander doesn't rebuild your trip — it makes one smart swap: "this hotel puts you €180 over. I found a similar boutique stay seven minutes away, bringing the total back to €2,290." Payments run through stripe, so no card details ever touch Wander itself.

Under the hood

The blueprint: Next.js, React, and TypeScript for the app. The OpenAI api as the brain — using structured outputs and function calling, so the model returns reliable itinerary data instead of free-form text. Amadeus for flights and hotels, Google places and routes for the map intelligence, a weather api for scheduling, stripe for payments, postgresql for trips and bookings, redis and background workers for price refreshes and long planning jobs.

I'm not a developer. But I can architect a product like this end to end — which apis, in what order, scoring on which criteria, where the AI decides and where it only verifies. That's the skill AI actually rewards: knowing exactly what to build and how the pieces think together.

Tall view of the Wander trip planning experience

Why I'm sharing the blueprint

Because this is what "working with AI" really looks like in 2026 — not asking a chatbot for recommendations, but orchestrating live data, real bookings, and personal taste into one product. And because I want you to see that designing something like this doesn't require an engineering degree. It requires curiosity, product thinking, and the willingness to learn how these tools connect.