Tray · for Northgate students
Live web build of the real app · the scan replays a real model output on a curated tray.
Tray matches your dining-hall photo to today's real menu. Accurate calories and protein in under ten seconds — no label scanning, no estimating, no food diary.
Watch the loop → A nudge fires as you reach The Commons. Snap your tray. Tray identifies the chicken, the beef & broccoli, the salad — pulls each item's published nutrition — and the only thing it estimates is portion, which you adjust in one tap. The ring fills. Done.
The vision model only decides which of today's known menu items are on your tray — a closed-set match against the live menu for your hall and meal. A small, solvable problem.
vision ML · the only "guess" about foodCalories and protein come straight from the dining hall's published nutrition database — never invented from pixels. If the menu says 290 cal, Tray says 290 cal.
menu database · the source of truthThe one thing Tray estimates is how much you took. It's shown up front, tagged "est.", and one tap changes it. The number is always yours to correct.
portion · the only estimatetop-1 accuracy over the popularity baseline. Restricting the match to today's menu is the whole trick — it turns open-world food recognition into a tractable, measurable problem.
held-out eval · numpy · $0 · loading…
Snap your tray. Tray identifies which menu items are on your plate, right from the photo.
Nutrition from the source. Numbers come from the published dining-hall database, not AI guesswork.
Adjust portions in one tap. We estimate how much; you confirm.
Three venues, every meal. The Commons, North Station, and Hillside Market — updated daily.
Rings, not spreadsheets. Calories and protein at a glance, zero data-entry.
Awareness, not obsession. Protein-forward, never shaming. Ten seconds, three times a day.