Product Build
An AI Journey Planner App
- Roles
- Product Design & Development
- Marketing
- Tech
- Client side: Vue, Capacitor, Swift
- Backend: FastAPI, Supabase, Resend
- Timeframe
- May 2026, Taipei, Tokyo
- Product @ Deejiar
The best journeys aren't planned. They're inherited.
Inherit, don’t generate
Taste can’t be generated. A local’s day is already proven — the model’s job is fitting it to you.
Deejiar empowers travelers to share the taste from real local tour’s journey
A generated itinerary is a guess; an inherited one is proof. Verified locals publish the days they actually live — you take one whole, then adjust it stop by stop until it’s yours.
How I built the AI development flow
Every generation drifts. So judgment sits at both ends and never in the middle: /plan defines intent from zero, /build executes the approved plan and may only stop and ask, /inspect checks what actually came out. Red-teaming lives in /inspect, not /plan — attacking a plan is speculation about speculation; attacking an audit is verification.
Feature build flow:
Defines intent. Writes no code.
Commands that support the build
Cuts a file to its smallest, highest-signal form — zero information loss.
How far did I move in two months
Ten releases in two months, v0.02.0 to v0.02.7 — the last one alone 110 commits in five days: a Butler thread that plans before it generates, explicit trial limits, and a Featured ranking smoothed so early journeys can’t sit on top forever.
Designing and building it myself is what makes that pace possible. There is no handoff, so the distance between deciding something and shipping it is an afternoon.
Where It Goes Next
Tokyo and Taipei today, with more of Asia rolling out. Guides join city by city, and when a guide covers your city, booking a private tour is one tap from the journey you already inherited. The harder problem stays the same: the more the model does for you, the more it has to be right about.
