About project
Fitness AI is a calorie tracker that turns one photo of your plate into numbers people can trust.
Product
Fitness AI
Category
Health and nutrition
Platform
iOS, light and dark
My role
Sole product designer
Timeline
6 weeks
The challenge: people quit trackers because logging is slow and AI numbers feel made up.
The result
9 s
to log a meal with one photo
4.1 / 5
trust in the AI estimates
6 of 8
found what was left today at a glance
The problem
Calorie tracking works. Most people just quit too early.
I started with one question: why do people give up? The same 3 answers kept coming back.
Client request
“If logging takes longer than eating, people stop.”
Sofia Martinez
Product Manager, Fitness AI
Problem 1 · Slow logging
Logging took too long
Typing every ingredient and guessing portions turned each meal into a form.
Problem 2 · Untrusted numbers
AI numbers felt like a guess
Photo trackers showed one number with no working, so people did not believe it.
Problem 3 · Wrong home screen
Home answered the wrong question
Weekly charts led the home screen. At lunch, people only need what is left today.
50%
of new users stop logging within the first week
Source: Fitness AI user research, 2026
Business task
Make logging fast enough to become a habit and honest enough to trust.
Decision 1 · Fixes slow logging
Photo first, typing as the fallback.
Decision 2 · Fixes untrusted numbers
AI is only useful when people can check its work.
Option A
One number
Fast, but there is no way to check it.
Option B
Manual entry
Accurate, but slow enough to make people quit.
Option C
Estimate with its working
Ingredients, grams and a confidence level, all editable.
Trade-off: the result takes more space than one number, so the total comes first and the ingredients sit below it.
Final design · Onboarding
10 questions, about 2 minutes, one personal plan.
Final design · Logging
6 ways to log, one sheet to choose from.
Final design · Tracking
3 tabs, 3 questions people ask every day.
Results
Logging became something you do between bites.
9 s
to log a meal with one photo
Fixes slow logging · 8 moderated sessions
4.1 / 5
trust in the AI estimates
Fixes untrusted numbers · post-test survey, 8 people
6 of 8
found what was left today at a glance
Fixes the wrong home screen · 5-second test, 8 people
Benchmark
Time to log one meal, same test
A leading manual tracker
Fitness AI
11 fewer taps from photo to logged meal.
What I learned
People trust AI when it shows its working. I now design every AI feature that way.