Guardian of the pass
The backstop on the line. It catches the wrong or incomplete plate before it reaches the guest. Watchful and protective, never a spy on the line.
LIVECamera-checked plates, every dinner service
Watches every plate
at the pass.
…and catches what’s wrong before it leaves the kitchen.
An overhead camera reads every plate, checks it against the live Toast ticket, and flags a mismatch the moment before it goes out. Silent when it’s right.
The pass is the counter where finished plates get one last human look before they go to the guest. AyeQ is the extra set of eyes standing there.
Fig. 01 · The pass · live capture
Live · Seneca /Source overhead camera /Runtime on-device /Link Toast POS
§01 The idea
Invisible intelligence
You see the outcome: the right plate. The AI is a quiet engine underneath, trusted because it’s trained on your menu, not a generic model.
Quietly confident
Clean, understated, precise. A high-end instrument that respects the craft of the kitchen, and earns its place at the pass.
§02 The problem
A wrong plate is expensive. Every single time.
During the rush, the expediter is only human. A missing sauce, the wrong side, a plate sent to the wrong table. Each one costs:
- 01A comped dish, and a remake on the fly
- 02A slower ticket for the whole table
- 03A one-star review that outlives the night
- 04A guest who quietly never comes back
- 05And the one nobody can afford: an allergen on a plate it should never be on
The pass is the last line of defense, and tonight it rests entirely on one busy person catching every error, every time.
§03 How it works
Four steps, every plate. In well under a second.
-
01
Watches every plate
An overhead camera above the expo line sees each plate the moment it lands, with no scanning or extra step for the kitchen.
-
02
Reads the dish & its parts
A two-stage on-device pipeline names the dish, then verifies its components (sauces, sides, garnishes) down to the plate.
-
03
Checks the live ticket
Each plate is matched against the open Toast order, so AyeQ knows what should be there, not just what is.
-
04
Flags the mismatch
Right plates pass in silence. A miss raises a quiet overlay: what was expected, what was seen, one tap to flag.
Fig. 02 · Component check
Fig. 03 · Ticket match
Modifier checked
§04 The catch, in action
The plate that was about to go out wrong.
The whole product in one moment. The ticket called for tomato salsa; the plate arrived without it. AyeQ raised the flag while the plate was still at the pass. Real footage, from a real dinner service.
Fig. 04 · The save · live service
§05 Allergens
The same eye that catches a missing sauce catches an allergen on the wrong plate.
AyeQ knows your menu, and what’s in each dish. When a plate carries something a guest flagged an allergy to, the expo sees it at the pass, live. The same eye, now watching for the one mistake nobody can afford. Running today at Seneca.
And it doesn’t guess at allergens. The vision probe detects the ingredients actually on the plate; each one resolves through a deterministic parser (the explicit, auditable ingredient-to-allergen map behind AyeQ’s allergy assistant), and is checked against what the ticket flagged. The detection is the model’s job; the allergen call itself is a fixed rule, not a prediction.
- Ingredient detected
- Deterministic allergen parser
- Matched to the ticket
- Flagged at the pass
Fig. 05 · Allergen flag
A rule, not a guess
§06 The technology
Why it can do this. On a standard iPad, right at the line.
On-device, always
Both models run on the iPad’s own silicon, the Neural Engine and the GPU, entirely on-device. Every live check happens right there, with no cloud round-trip, nothing to wait on mid-rush, and nothing streamed out of your kitchen in real time.
A model per restaurant
Each kitchen runs its own model, fine-tuned to its exact menu and plating. It sharpens over time from your own service; the only thing that ever leaves the iPad is labeled training crops, kept in storage private to your restaurant and never pooled with anyone else’s.
Two models, one plate
A fast detector finds every dish on the pass in a few milliseconds; a second model then reads its components down to the plate: sauces, sides, garnishes. Two specialists, both on-device, with nothing to post-process.
Plugged into Toast
AyeQ reads the live ticket straight from your POS, so the check is against the real order, in real time, not a guess about what the plate should be.
Instrument readout
- 0.95
- mAP50 · detection accuracy on Seneca’s live menu
- ~13ms
- per detection, on the iPad Neural Engine
- 2-stage
- on-device CoreML pipeline · detector then probe
- 75
- components AyeQ can recognize on a plate
- Real-time
- matched against the live Toast ticket as it’s plated
- On-device
- every check runs on the iPad itself
06·1
06·2
06·3
06·4
06·5
06·6
Measured on Seneca’s live menu and hardware; on-device inference on the live check. Service-accuracy figures (plate-to-ticket match) are reported separately, with their method and false-alarm rate.
§07 The operator
The kitchen behind it.
AyeQ runs at Seneca every dinner service: an overhead camera at the pass, checking every plate against its ticket. It has been live for months, sharpened against a real rush the whole time.
Not a pilot staged for a demo. The working kitchen it earns its keep in.
§08 Honest about what it is
What AyeQ is. And what it isn’t.
AyeQ is
- A check on the plates that slip past in the rush
- Trained on your exact menu and plating
- Honest about its false alarms
- Private to your restaurant; your data is never pooled with anyone else’s
AyeQ isn’t
- A replacement for your expediter
- A generic, one-size-fits-all model
- A claim of perfection
- A live video feed streaming your kitchen to the cloud
Come stand at the pass.
The best demo is a dinner rush. Get in touch, and watch AyeQ check real plates against real tickets, live at a working restaurant.