Behind the build

How NutriFitr recognizes a meal from one photo

June 18, 2026 · The Airus Solutions team

Every calorie tracker dies the same death. Week one, you weigh your rice. Week two, you search "chicken curry" and pick the third result. Week three, you stop. The problem was never motivation — it's that manual logging costs more effort than the insight is worth.

NutriFitr's answer is a camera. You photograph your plate, and a few seconds later the meal is logged with calories and macros. Here's how that pipeline actually works — and the decisions that shaped it.

From pixels to a plate

When you snap a photo, NutriFitr's AI does three jobs in sequence: it identifies the foods on the plate, estimates the portions from visual cues like plate size and food volume, and maps each item to nutrition data to produce calories, protein, carbs and fat. A mixed plate — say rice, dal and a salad — comes back as separate line items you can adjust individually, not one mystery blob.

Estimates, honestly

No model can weigh your food through a lens, so we don't pretend otherwise. NutriFitr treats every result as a fast first draft: portions are one tap to nudge up or down, and any item can be swapped or removed. In practice the draft is close enough that most logs need zero edits — and a log that's 90% accurate and actually happens beats a perfect log you quit doing.

The design target was never "perfect nutrition data." It was "logged in under ten seconds."

It all flows into Apple Health

Everything you log syncs to Apple Health, so your nutrition sits next to your steps, workouts and weight — and works with whatever else you use. Your photos are used to analyze your meal, not to build an advertising profile; see our privacy policy for the details.

What we learned

The AI is the headline, but the retention came from friction removal everywhere else: launching straight into the camera, collapsing the review step to one screen, and never making you name a meal at 8am. If you're building an AI feature into your own product, that's the transferable lesson — the model gets you a demo, the interaction design gets you a habit.