Case Study · Personalized Nutrition and Meal Planning
Kodexo Labs built Nutrition Buddy, an app that asks a person a handful of questions, works out how much they should eat and in what proportions, and hands back real meals with portion sizes—with every gram checked against verified food records rather than guessed.
01 · Overview
Building a personalized, macro-aligned meal plan is normally manual and error-prone: working out daily calorie burn, splitting it into protein, carbs, and fat, and then finding food that adds up to those numbers by hand.
Nutrition Buddy automates that arithmetic. It asks for age, height, weight, sex, activity level, dietary preferences, and how many meals a day the person wants, then generates a meal plan with portion sizes and ingredients, checking the result against verified nutrition data and re-fitting portions until the numbers hit target.
02 · The Problem
Each one comes straight out of the specification for this build, and each one is a moment a person quietly gives up on eating better.
Working out daily calorie burn and splitting it across protein, carbs, and fat is a manual calculation, done by hand, per person.
The Mifflin-St Jeor calculation and macro split run automatically the moment someone finishes onboarding.
Generated meal plans often lean on niche ingredients that a person cannot find or will not cook.
Meals are built from common ingredients like chicken, rice, and potatoes, with recipes rotated so week two looks different from week one.
A tool built for one way of eating turns away everyone who eats differently.
Six dietary categories are supported, including omnivore, keto, vegetarian, vegan, gluten-free, and dairy-free, and excluded ingredients never appear in a meal.
Setting an unrealistic goal, like losing a lot of weight in a short time, can go unchallenged and quietly become unsafe advice.
Goal timelines are checked as they're entered, and a timeline that isn't realistic gets flagged before a plan is built around it.
03 · How It Works
Here is the same journey a person's answers take every time, step by step, in plain terms.
Age, height, weight, activity level, dietary preferences, weight goal, and how many meals a day you want.
It calculates the calories your body burns in a day, adjusts that number for your goal, and splits the result into grams of protein, carbs, fat, and fibre.
Using common ingredients, checking every calorie and gram against verified nutrition data, so nothing is a guess.
Portions are adjusted and rechecked up to five times until the plan actually hits your daily target.
A full day of meals with portions, or a clear message that the goal isn't realistic yet—never a plan that quietly doesn't add up.
04 · The Solution
Each screen asks for one thing or shows one thing. Together they take a person from a blank start to food they can buy on the way home.
No account to create and no password to invent. A name and an email, and the plan is already tied to a person the app can send it to.
A progress bar on every step, so nobody wonders how much is left
You are taken straight to the next question, with no extra setup
Biological sex, height, age, weight, and how active a day actually is. Activity is described in real terms instead of asking people to rate themselves.
Ranges checked as they type; adults only
Kilos or pounds, and a goal timeline that gets refused if it is unsafe
Eight tiles cover the way people actually eat, and anything ruled out here never appears in a meal. Then one to six meals a day decides how the calories get split.
Allergens can be excluded outright, so they never turn up in a portion
Six dietary categories covered, reaching more people than a single-diet tool
Before a single meal is generated, the app lays out what it worked out: calories burned at rest and across a whole day, the goal adjustment, and the daily target split into macros.
Every figure shown, not hidden behind one calorie number
Each macro links to the food choices behind it
This step is optional. For anyone who cares, each macro gets its own tab, its own set of sources, and a running gram total against the day's target.
Six sources under each of protein, carbs, fat, and fibre
Change one share and the rest redistribute automatically
One card per meal: what to eat, how much of it, the calorie total, and the protein and fat it contributes. Recipes rotate too, so week two does not look like week one.
Taken or not taken, marked per meal, so the day becomes a record
A feedback box on every meal, and the next plan reflects what was said
05 · Outcomes
Every figure is modelled from the specified feature set and capped adjustment loop, and each traces back to the specification.
The model composes meals; it does not supply the numbers. Calories and grams come from USDA FoodData Central and Cronometer, and every plan is checked against those records before anyone sees it.
Twenty minutes with the team that shipped this. Bring the calculation you cannot get wrong, and leave with a straight answer on what it takes to build around it.
Related work
