DESIGNING & SHIPPING
A 0→1 Mobile App that turns your Food Photos into a Culinary Journal
This is a solo passion project in which I designed and shipped a live AI product end-to-end, learning how to hone my critical thinking skills to prompt and use AI agents efficiently.
THE IMPACT
Live
Deployed PWA
Multi-user
Secure auth
AI-powered
Sonnet vision
THE PROBLEM
AI tools are more capable than ever, yet their performance still boils down to how well a designer can prompt, steer, and decide what to hand over.
IS DESIGN REALLY "DEAD?"
THE GOAL
Vibe-code and ship a real, multi-user, secure product by directing AI tools, and be able to defend every meaningful decision as my own.
WHAT SHOULD I BUILD?
I wanted to build something meaningful out of data that already existed.
I take a lot of photos of my food. I do not always post them anywhere, but I am a very visual person, and I like scrolling back through my camera roll every now and then because those photos feel like little memories.
I wanted to do something fun and genuinely meaningful with all of them, and AI workflows and automation were what made that feel low-effort enough to actually be worth it. I kept thinking about how iOS Photos resurfaces memories on its own, and when I combined that with the joy of a Spotify Wrapped, the idea of a culinary journal started to take shape…
Camera Roll of endless food photos
Spotify-Wrapped style insights about your food personality, vibe, and top cuisines!
COMPETITIVE ANALYSIS
No competitor currently combines AI-powered cuisine recognition, with identity reflection, and a wrapped-style narrative.
Feature
Atemate
Treatly
Savor
Yummi
Memolli
AI-powered food & cuisine identification
Health-focused tracking
Rating food/restaurant tracking
Culinary identity based on food vibe & palette
~
~
~
Cuisine based World Exploration
Spotify-wrapped style narrative
Yes
No
~
Partial
Food Wrapped moves furthest toward identity and discovery: it turns meal photos into a playful self-portrait, not just a food log.
THE PRODUCT
Food Wrapped uses AI vision to turn your food photos into a culinary journal with insights on your Food Personality, Top Cuisines, & World Exploration!
Focus on the memory, not logistics!
You snap or upload a meal,
An AI vision model autodetects and fills the dish name, cuisine, vibe, and other metadata like date and location for you, so there are no forms to fill in and nothing to tag manually.
Optionally, you can add a memory to remember that meal by!
Over time, the app builds a picture of your culinary identity:
your top cuisines
your evolving food personality and
the regions explored on a world map!
Your Food, Wrapped!
Whenever you want, the app generates a Spotify Wrapped style recap of your food experiences for any period of time that you can then share on your socials!
The app is a celebration of food as memory and adventure, and it is deliberately not a calorie counter or a restaurant rating tool.
There is no nutrition tracking, no rating, and no restaurant logging, because I wanted the experience to feel simple and like looking back on your experiences with pride rather than auditing them with guilt.
THE AI TOOLS I USED AND WHY
I treated each AI tool like a knife in a kitchen, reaching for the most appropriate one for each specific cut.
Claude, as a thought partner and for the PRD
Before touching a vibe-coding agent like Cursor, I used Claude Chat to fully flesh out the idea and scope the project into a PRD. The clearer and more complete my thinking was going in, the better the AI could execute it, so I wanted to get the intent right before handing anything over for the AI agent to guess.
Framework adapted from Tina Huang, "Vibe Coding Fundamentals in 33 Minutes" (2025).
How was it useful?
I found Claude's knowledge of the competitive landscape helpful in rapidly pressure-testing my idea from multiple angles before designing.
Figma Make, for visual direction
I put together some design guidelines, and accessibility rules using Claude Chat, before adding them to Figma Make to explore and lock a visual direction.
To get sharper output, I applied the TC-EBC framework using a custom, publicly-available GPT Make Prompt Assistant.
Task: Design a Home screen that motivates users to explore and diversify cuisines through their food journal.
Context: This is the primary landing screen users see after opening the app to review their food journey.
Elements: Include key stats, cuisine collection, achievement progress, and a Wrapped teaser.
Behavior: Prioritize glanceable insights and gamified progression while encouraging users to discover new cuisines.
Constraints: Use the existing Food Wrapped design system, light theme, orange primary color, mobile-first layout, and avoid gradients.
Example of TC-EBC framework for the first Home screen draft
How was it useful?
Figma Make's functional prototype made UX decision-making easier by helping me think of edge-cases faster than would have been possible using a manually constructed happy path prototype.
Cursor, for the actual build
I treated Figma Make's output along with the PRD as a spec for how the app should look, feel, and behave and had Cursor build the actual function underneath it. The prototype showed the intent, and Cursor enabled me to link it to a real database and deploy it.
How was it useful?
Vibe-coding in Cursor closed the gap between designing and testing, so I could catch and fix real usability problems in the moment instead of discovering them much later.
Cursor enabled me to fine-tune app interactions using real data, such as the horizontal scroll for top cuisine entries
Supabase, Vercel, and the Claude Sonnet API
Supabase handled the database and user authentication.
Vercel handled deployment via GitHub, where I committed continuously for version control.
Claude Sonnet ran the food image vision classification via an API
This streamlined the process of users taking a meal photo and the system returning structured data such as the dish name, the cuisine, a confidence score, and whether or not a person was detected in the image for privacy reasons.
DESIGN DECISIONS
The AI handled the grunt work, but these were the decisions I owned.
Surfacing privacy without adding friction
Photos are sensitive data. I did not want users worrying about accidentally uploading personal non-food photos to the app. This was particularly crucial for the bulk upload option where users can manually import food photos from their camera roll.
The AI vision model flags when a human is detected in an upload, and the user gets a gentle heads up to keep or discard that entry right then and there.
Caution is surfaced with ease, exactly when it is relevant, and without any extra settings to manage.
Choosing honest metrics over impressive sounding ones
AI often introduces made-up fancy stats like "you're in the top 5% of foodies" which can be confusing and difficult to make sense of. To simplify the "World Explorer Rank" and prevent overwhelm, I instead chose to calculate them as a percentage of 35 popular cuisines so users know what they are working towards. Each metric also comes with a short explanation of how it is calculated and how users can move it.
A more informed user has more agency, and more agency builds more trust.
Every stat in the app is presented clearly using this logic.
I also tuned the points system to reward breadth over volume, so trying a new cuisine is worth more than logging the same dish again. I wanted the incentive to match the soul of the product, which is exploration rather than grinding.
The information architecture tells a story
Cursor's unguided output was generic, single screen, and flat. I owned the hierarchy of elements on the screen to elicit emotions of pride and nostalgia.
Photos take priority over numbers because food carries emotions & memories!
The meal carousel, made up of the user's own photos and therefore their own memories, is the first thing they see when they open the app!
WHAT THIS LOOKED LIKE IN PRACTICE
I do the judgment, AI does the grunt work
THE GOAL
A Snapchat style world map where users could zoom across regions and see, at a glance, the breadth and depth of where their palate had travelled.
MY ROLE
Deciding what the map should portray, how the zoom-to-explore interaction should feel, what belonged on the summary card, and how users can be encouraged to explore more regions.
AI's ROLE
Assembling the actual map view for me to test and play around with so that I did not have to build it from scratch.
A World Map to see the breadth and depth of where your palate has traveled!
WHAT I DELIBERATELY DID NOT BUILD (YET)
In an ideal version, the app would link to your camera roll and auto-detect food photos as you take them, turning everyday data into storytelling with no manual upload at all.
This, however, opens up real security and iOS authentication challenges that were genuinely out of scope for this build. So I made the disciplined call to defer it with privacy front of mind.
REFLECTIONS
The designer's job is becoming more fluid with engineering, and I think that could be powerful!
Having an AI build the design lets me explore edge cases and see my own vision come to life far sooner, which means I get to spend more of my time on the logic of an experience and less on the mechanics of displaying it. It has made me a faster collaborator and a more critical one, quicker to test a decision and less precious about holding onto one.
The better you are able to prompt an agent, the more satisfied you are with its performance.
What does not change is the part that was always the real job. Thinking from a human-centric perspective, anticipating the behaviours and emotions of the people who will actually use the product, and making the call about what matters most in a given context. The tools will keep advancing, but knowing what is worth building, and staying accountable at every consequential decision, is the part that stays mine.
What I would do Differently
User Research
If I were to do this project again, I would talk to potential users, or run a social media survey to learn more about the needs and perspectives of potential users so I can design more deliberately for them.
Thanks for sticking till the end! :)
If you would like to get access to the deployed app, please feel free to email me!
tariq.memuna@gmail.com