Four conversations, one kitchen
Kitchen Sora is a fictional izakaya with a takeout counter. Tap a scenario to play it, tap the cards and chips inside the phone, or type your own order — “2 chicken curry, no onions, 18:30” is the one the parser was built for.
The right-hand panel is the shop’s side of the same conversation: four KPI tiles and the event log the kitchen would actually see. Nothing here reaches a real kitchen.
What each scenario is showing you
15:00, twenty portions cut and nothing sold. One push to the 62 guests who have ordered grilled fish in the last 90 days — not to everybody, because a push costs money and an untargeted one costs goodwill as well. The order is taken, paid and ticketed before the doors open.
Quantity, size, a modifier and a time, in the order a guest would actually say them. The bot reads it back as a confirmation card and changes nothing in the kitchen until the guest agrees with what it understood. “NO ONION ×2” then prints on its own line of the ticket.
Two versions of one dish with two different answers. The bot gives both from the chef’s notes, refuses to call the kitchen peanut-free, and escalates the cross-contact question to a named chef. Then it offers the dishes whose notes are clear, and flags the allergy on the ticket.
18:45 holds four boxes and four are taken. The bot says so, explains why when asked, and offers the next two slots with room. An order the pass cannot cook is worth less than an order forty minutes late, and this is the setting that enforces that.