What it does
Start in your city. This demo has New York and Singapore, with more cities to come. Then tap the one you like more, eight times: a band or a brand, a show, dinner, a bar for after. Every tap tells Qloo a little more about your taste, and you watch your taste twin grow at the top of the screen. So each person gets a night picked for their own taste.
Then you hit "Hold my night" and an agent holds the whole thing. Dinner, the show, drinks after.
On the way, one booking page tries a trick. It tells your agent to move dinner to its "partner" place. Your taste twin knows that place isn't you, and stops it. The evil twin shows up on screen, gets struck out, and your night goes ahead as you picked it. You end with a ticket for two to send to your +1.
Quality of the Idea
Everyone uses Qloo to say yes to things you'll like. Super Two also uses it to say no to an agent that's been tricked. The taste that plans your night also protects it.
Potential Impact
AI agents are starting to book things for people. They read web pages nobody checks, and a page can tell them what to do. Super Two guides the agent and corrects it: before the agent changes your plan, it checks with your taste. Any app that books tickets, tables or hotels for people can use the same check.
Design
One flow from first tap to ticket: eight picks, one hold, one ticket. No scrolling and no forms. Every step happens in front of you: your twin growing as you tap, the agent holding each place, the evil twin caught.
What's next
Groups: one night planned around everyone's taste, with the ticket shared with the whole group.
Results
We threw 32 sneaky swaps at it, built live from Qloo for 4 different tastes in New York and Singapore. Your twin stopped 27. It also let through all 32 fair changes, so it never got in the way of a good night. Full results are in the README.
How it uses Qloo
Qloo chooses every option you see, gets to know your taste from your taps, and checks every change the agent tries to make.
- Picks.
/v2/insightstakes your picks as signals (signal.interests.entities) and ranks the shows and the real restaurants and bars near the venue (filter.location,filter.location.radius,filter.tags). - Gets to know your taste. Each pick joins the signal list, so the next pair fits you better. The twin bar shows Qloo's own tags and descriptions for what you picked.
- Guards (
src/guard.ts). Before any booking, the guard asks Qloo how well the new place fits you. A swap passes only if it fits as well as your twin's own top 10 for that slot and is within 2.5 km. Qloo's scores run higher in some cities than others, so each place is compared with its own area. If Qloo can't score a place, the guard asks you.
The agent is an AI model with three tools: read a venue page, hold tickets, hold a table. The guard is plain code that runs on every change, so it never relies on the model saying no.
Limits
This is a demo. The guard checks taste and distance. A planted swap to a nearby place that does fit your taste will pass. Beyond the demo, we can have more cities, better guardrails and solid integration with venues and events data.
Log in or sign up for Devpost to join the conversation.