Point a camera at a plate and you get a grid of pixels. Turning that into “420 kcal” takes three distinct steps, and each one can be wrong in its own way. Here's what runs, in order.
1. Identify the dish
The first pass asks a simple question: what is this? A vision model proposes candidate dishes and scores its confidence in each. A bowl of greens with sliced chicken and avocado is a much narrower search space than “food”, and narrowing it early is what keeps the rest fast.
2. Split it into ingredients
A dish name alone isn't enough — two chicken salads can differ by 300 calories depending on dressing and portion. So the second pass segments the image into components and estimates the mass of each. This is where most of the uncertainty lives, and it's why we show a per-ingredient breakdown instead of a single number you have to trust blindly.
3. Look up the nutrition
Each identified ingredient is matched against a nutrition database and scaled by its estimated mass. Calories, protein, carbs and fat are summed, and the result carries a confidence score derived from how certain the first two steps were.
That confidence score is the honest part. When MealLens is unsure, it tells you, and you can correct the portion in a tap. A tracker that pretends to certainty it doesn't have is worse than one that admits the range.
Try it yourself
Point your camera at your next meal.