How a photo becomes a nutrient profile.

1. Image recognition

When you photograph a meal, a vision model identifies the ingredients present — typically 1–8 components per dish. It returns a labeled ingredient list with a confidence score for each item.

For mixed dishes (curries, stews, smoothies) the model uses both the visual and any user-provided context (e.g. a short caption) to break the dish into its likely ingredient set.

2. Portion estimation

Each identified ingredient is assigned a portion estimate in grams, derived from plate geometry, reference objects in the frame (when available), and prior distributions for that food type. Portion estimation is the single largest source of error in any photo-based tracker. You can manually correct portions in the app.

3. Food-composition mapping

Each ingredient + portion is mapped to a verified food-composition database to produce calories, macronutrients, and 25+ micronutrient values per meal. We prefer government and academic sources over crowdsourced databases:

• USDA FoodData Central (United States)
• Livsmedelsverket (Sweden)
• EFSA Food Composition Database (EU)
• FoodComEx and Frida (Denmark) where region-specific values matter

4. The 14-day rolling window

Each nutrient value is added to a rolling 14-day average. We chose 14 days because most water-soluble vitamins replete or deplete on that horizon, and many fat-soluble nutrients show meaningful change on a two-week scale.

This window also makes the app forgiving: a single low-magnesium day doesn't trigger an alert. A two-week pattern does.

5. Glycemic balance score

Each meal receives a 0–100 glycemic balance score derived from its carbohydrate quality (fiber, glycemic index of components), protein and fat content, and meal composition. This is an estimate of glycemic impact, not a measurement of your blood glucose. We do not integrate with CGMs at launch.

6. Recommendations

Recommendations are generated from your gap profile. We select whole-food sources first (e.g. sunflower seeds for magnesium), tag them by the nutrients they close, and note absorption co-factors where relevant (e.g. beta-carotene with fat).

Limitations

See our accuracy page for honest error ranges, known failure modes, and what we are doing to reduce them.