← Jiaqi (Evey) Gou Case 03 / 04 · Independent
AI product · Consumer · Multi-user

From AI itineraries to group decisions

Every AI tool can write an itinerary. The hard part of a group trip is getting four friends to agree on one. Trip Together is built around that moment.

Try the demo ↗ Seeded Lisbon trip with 4 friends
Role
Product owner & builder
Users
Friend groups — an organizer plus invitees
When
Fall 2026
Status
Public demo · end-to-end QA in progress
Trip page for a Lisbon demo trip: invite link and the four friends who are ready, each with their vibes
01Crew — one invite link, and who's ready with which vibes
Preference step: tappable vibe cards such as great food, hidden gems, culture and nightlife
02Vibes — tap cards, then add pace, notes or voice
AI plan with a group harmony score, a flagged vibe clash with a proposed compromise, and vote buttons on each option
03Vote & reconcile — AI flags the clash and proposes a compromise
01 — The reframe

It started as a solo AI travel assistant, grounded in a strategy memo I wrote on AI travel concierges. Group trips were explicitly out of scope for v1.

Then I looked at where trips actually stall: in the group chat. Preferences scatter, someone stays silent, nobody wants to be the one who says no. Generating options is cheap; reaching agreement is the unsolved problem. So I changed the scope.

02 — The flow
01
Create
Trip, place, rough dates
02
Invite
Link, no account
03
Prefer
Vibe cards, pace, notes, voice
04
Generate
2–3 AI options per slot
05
Vote
Yes / fine / definitely no
06 · core
Reconcile
Conflicts, compromises, ties
07
Lock
Timeline, .ics, story card
03 — Designing disagreement

Reconciliation is the differentiator, so I designed for the messy cases a group actually hits:

Veto
A "definitely no" becomes an open issue with the reason attached — not silently outvoted.
Tie
AI drafts a compromise; the organizer can regenerate with specific guidance.
Absent friend
Proceed now, majority lock, coin flip, or a "go-with-the-flow" persona stands in.
Override
The organizer keeps the final say — AI facilitates, it never makes an irreversible call.
04 — Product calls
Options + votes, not one answer
Showing alternatives and asking for explicit votes keeps people in control and makes disagreement visible early.
Structured cards + free notes
Cards are fast to fill; notes and voice capture real constraints. AI turns notes into chips the user confirms.
No-account invites
A per-trip link and browser token instead of sign-up — friends join in seconds; access control is the trade-off I'm still testing.
Versioned itineraries
Each plan is stored as a structured version, so regeneration, voting and export don't overwrite each other.
05 — Under the hood
Data
Trips, travelers, preferences, itinerary versions and votes, with realtime sync
AI jobs
Preference extraction, generation, regeneration, reconciliation, absent-friend persona
Platform
Lovable Cloud + AI Gateway; voice input via speech-to-text
06 — Honest status

The demo is public. Before I call it done, I'm validating the full flow with multiple real travelers: realtime voting, conflict handling, AI failure cases, invite-link access and exports. No adoption claims until there's data.

The lesson so far: in a multi-user AI product, the model is the easy part. The product is how people disagree.

Next case · 04 Meal Gacha! →