Garcon began with a very familiar problem: organising food orders for a family evening.
Everyone sends their choices in a group chat. Some people change their minds, dish names do not quite match the menu, messages get buried, quantities are duplicated, and one person has to reconstruct the final order manually. When the food arrives, somebody inevitably asks:
“Where’s mine?”
“This isn’t what I ordered.”
That was the starting point for Garcon - Your Takeaway Order Copilot, which I built for OpenAI Build Week during the final stages of the 2026 World Cup.
Garcon lets an organiser review a menu, create an ordering event and share one secure link. Guests do not need accounts: they enter a name, make their selections, add an optional note and submit. The organiser receives both a per-person distribution list and aggregated quantities, with the totals reconciled between the two views.
Once I had the food-ordering flow working, I started noticing the same problem elsewhere. For example:
- Preparing lunch or refreshments for an event at work.
- Collecting choices for a school activity.
- Planning food for a birthday or family gathering.
- Estimating portions for homemade dishes, desserts or a potluck.
Garcon is still food-first. But the same small problem appears elsewhere: a group is choosing from a fixed list, and one person needs a dependable answer about who wants what and how many are needed.
I have not built the non-food versions yet, but I can imagine the same approach helping with workshop choices, school-event packs, clothing sizes or volunteer tasks. That feels like a natural direction for the idea rather than something the current application already claims to support.
Not every group decision needs a complex form builder. It should not require a spreadsheet expert either. And it definitely should not have to be reconstructed from multiple chat messages.
I built Garcon with React and Supabase, using ChatGPT to challenge and refine the idea and Codex with GPT-5.6 to implement it. Codex was particularly useful when working through the less visible details: securing the no-account guest flow, keeping prices on the server, testing submissions and deploying the application. ChatGPT Remote also let me continue resolving last-minute issues while I was away from my desk.
The result is live and usable: an organiser can create an event, share a no-login link, and see guest responses in both per-person and aggregated views. Behind the scenes, the menu is frozen for the event and prices are calculated on the server.
It may sound like a small problem. It is. But small logistical problems have a habit of taking over an otherwise enjoyable evening. If Garcon removes one of them, that is already useful.
OpenAI Build Week entry:
https://devpost.com/software/garcon-your-takeaway-copilot-4k8aot
Try the application:
https://garcon-ordering.danuw.chatgpt.site
Corrected demo video:
https://www.youtube.com/watch?v=8BzQ_Pnz48w
I would love feedback on where this idea could go next—especially other situations where collecting a group’s preferences becomes more complicated than it should be.
Top comments (0)