Simple – AI food tracking
11 Apr, 2022
As the first designer at Palta-backed Alms, I took the product from zero to first release in five months, led redesign through two pivots, and owned everything from onboarding and core flows to design system, brand and landings.
Timeline
2020–2022
Role
Founding Designer
Scope
UX/UI, Design System, Brand
Product
0
1
, B2C, Wellness
Surfaces
iOS, Landing
Media
Alms was a social-impact wellness app funded by Palta (Flo Health, Simple, Lensa, etc). It helped people feel better through small real life actions, like helping others or building healthier habits.
Over two years it moved through three models: personal challenges, social network, and creator-led platform. The audience was 30–40+ already paying for apps like Headspace and Calm, and the final direction used creators as the distribution layer to bring them in at scale.
Simple started as a fasting timer — strong category leader, 15M+ downloads, narrow product. The growth question was clear: what comes after the fast?
Food logging was the obvious next layer. It could anchor daily engagement, generate the personalization data the rest of the product needed, and build nutrition credibility that fasting alone couldn't carry. By Q1 2024, the company-level key result was explicit: food feedback had to be ready to drive a new habit by Q2.
But Simple's users weren't calorie counters. They didn't want MyFitnessPal-style admin. They wanted a healthier version of how they already ate. The brief sat between two things that don't usually go together: make food logging useful enough to matter, light enough that people actually keep doing it.
Exploration
The food database was the bottleneck. UXR was blunt about it: users couldn't find what they ate, search lacked autocorrect, and the act of "building a meal" reinforced the feeling that tracking was tedious and not for them.
I explored the system around it — daily progress, meal timelines, nutrition tips, post-log recommendations, lightweight feedback moments. Different shapes of the same question: could the loop become valuable enough that input friction stopped killing the habit?
Eighteen months of small, reactive experiments moved retention by only 1–2%. The exploration isn't a trophy wall. It's a snapshot of a team learning that incremental fixes to a database-first model weren't going to be enough.
Making the input effortless
By early 2023, AI was good enough for a different bet. Instead of trying to build a better food database than MyFitnessPal, we decided to stop making users think like a database at all.
The new noting tracker I designed let people type or speak what they ate in normal language — then turned that into a structured entry the in-house feedback system could actually score. UXR landed where the design did: "Users perceived taking notes as easy and simple way of logging." A sentence-starter in the input field made the moment feel even lighter.
Making food tracking worth it
The noting tracker solved input. It didn't solve the reward. After logging, users hit a wall of macros and nutrient bars — technically correct, practically useless. UXR landed the point clearly: "When it comes to food you want to have a feeling of comfort." People didn't want to calculate. They wanted to know how the meal landed.
We took the European nutrition label system as a reference point and designed a four-tier meal score: Low, Fair, Good, Optimal. I worked with a graphic designer to develop flat illustrations that carried the tone the research pointed toward — warm, encouraging, non-shaming. It turned a nutrition result into something users were willing to look at.
The surface won the British Dietetic Association Digital Innovation Award 2023/24 for Simple App: Nutrition Scores and Food Feedback — external recognition, from a body that knows nutrition, that this was a genuinely useful approach.
Making the next step obvious
Once Food Feedback was in daily use, a new problem emerged. The character made nutrition feedback approachable, but it was taking space the nutrition result needed. Users coming back day after day were asking a more functional question: what did I score, why, and what do I do next? Specifics had to land in seconds, not just be implied by a colour.
The trigger was internal and behavioural: the character presentation was making feedback harder to scan at speed and, for returning users, starting to feel less reliable as a nutrition signal. The design task was clear — move the score to the top of the hierarchy. Result first. Reasoning underneath. Next action close. Personality steps back from the headline.
By this point I was also owning the AI Coach direction, and I brought that work into Feedback: AI-generated explanations replaced static copy, making the reasoning under the score more specific and useful. I collaborated with another designer on sharpening the four-tier scoring system into a more glanceable read. The judgement lands fast. The AI explanation follows. Nothing competes with the result before you've read it.
The experiment validated the direction. D3 meal-track retention improved +3.6% for new users. Time on the result screen dropped 10%. Users spent less time on the result — not because they disengaged, but because they got the answer faster.
Making food tracking multimodal
By this point, the feedback loop worked. The remaining friction was the input itself — users still had to describe or type what they ate. Avo Vision removed that. Point the camera at a menu, a plate, or a dish, and the app does the rest.
At a restaurant, scan the menu. The app reads each dish, scores it, and surfaces a best pick. You shouldn't have to remember what you ordered at Cecconi's to log it later. It shipped with real limits — dense menus, small print — but the core problem was solved.
At home, point the camera at your plate. The app identifies the components, scores the meal, and if you want to go further, hands off to Coach Avo directly. Two taps from photo to logged meal.
I liked working on this enough to keep exploring in my own time. One idea I kept coming back to as AI models improved: collapse the separate scan modes into one, and surface specific nutrition feedback as you scan — not after, not in chat. We didn't ship this, but it was one of those moments where the work felt worth it on its own.
Avo Vision moved the product retention +2.1% and Avo retention 3d+ moved +4.5%, both significant. By May 2025, 15–20% of all daily meal logs came through the camera. MedTech Breakthrough named smart camera functionality as one of Avo's core pillars when Simple won Best Virtual Coach in 2025 and again in 2026.
Food logging had been an admin task. This is the version where it stopped feeling like one.
Conclusion
Tracker retention
+34%
Cumulative
Product retention
+12%
Cumulative
British Dietetic Association
Innovation Award
Nutrition Score and Feedback
Acknowledgements
Built with love from Minsk with Alex Nevedovsky, Sasha Khadeka, Nick Shchetko, Sofia Chop, Joanna Buchmeyer, Roman Kutanov, Andrei Lunevich, Sergei Borovkov, and Dzianis Nikitsin. Thanks to everyone who cared, tested, and gave advice along the way. Alex – thank you for seeing my potential early.