Vibecoding Cooking app

COMPANY

Side project

mY ROLE

Forward Deployed
Designer

team

Me, Claude, Antigravity
Vercel

timeline

July-Nov 2025

Challenge

Most meal planning apps are built for hyper-organized planners. They require spreadsheets, calorie tracking, and rigid schedules. But most young professionals aren't rigid planners.

They are tired people standing in front of an open fridge at 7 PM wanting to be told what to eat.

"What's Cooking?" is a mobile-first web app designed to act as a trusted kitchen companion. It bridges the gap between discovering recipes (the Instagram graveyard) and actually cooking them. It features an intelligent "Surprise Me" engine, pantry-aware filtering, and frictionless meal planning.

My Role

As the founding product designer, I owned the end-to-end design of this product from zero to launch and beyond.

I worked alongside 2 engineers and a UX researcher, collaborating directly with the founding team to shape both product vision and execution. My role stretched across three modes: designer, research co-pilot, and workshop facilitator.

This was the first time I operated without a design system, a defined brief, or a team structure to fall back on and it taught me more about product design than any role before it.

250K

Installs within 8 months · 15K MAU

32%

Trial start to trial conversion rate

4.9 stars

Rating on App store + Apple feature nomination

I need to stop overthinking dinner

I want to be told exactly what to make based on what's in my kitchen, so I can save my mental energy after a long workday and just start cooking.

I need to actually cook what I save

I want to quickly pick tomorrow's lunch and know if I need to prep anything tonight, so I can go to sleep with complete peace of mind.

I need to feel sorted for tomorrow

I want to quickly pick tomorrow's lunch and know if I need to prep anything tonight, so I can go to sleep with complete peace of mind.

Jobs to be Done:

How

might

we

build a

meal system

that adapts

to chaos

than

help

demanding

disclipline?

System to Context

I start in the data looking at drop-offs, session patterns, competitor gaps and build a structured audit before opening Figma. The Trust Lever Framework made that repeatable at e-commerce brand.

To Co-create and Ship

I wanted to understand what it actually feels like to build so I deployed my first web cooking app using Claude Code and Antigravity, without writing a line of code from scratch and learning how AI learns from context faster.

To Co-create and Ship

Research Approach

Qualitative interiews: Along with the user researcher I'd accompany to analyse qualitative interviews and sense.make on Dovetail.

Qualitative interiews: Along with the user researcher I’d accompany to analyse qualitative interviews and sense.make on Dovetail.

Moderated interviews: For early problem discovery we would pitch potential customers three competitor products to learn from what is currently working for them

Moderated interviews: For early problem discovery we would pitch potential customers three competitor products to learn from what is currently working for them

Key Metrics my team owned:

Activation: % of new families who complete onboarding + log first sleep within 48h→ From onboarding to first 7 days trial journey

Conversion: Trial start rate to trial conversion by having felt at least one successful sleep routine. CTR on 'Subscribe now' → Paywall design

Retention: Monthly active paid users, 60 Day Customer Retention, Subscription repeat purchase rate.

Empathy driven UX Copy

When I noticed that 90% of our active users were moms, I pushed for a shift in how we thought about the product audience. I proposed a small but deliberate addition: an affirmation feature designed specifically around the emotional reality of motherhood. A signal that the product saw her, not just her baby's sleep schedule.

The founders backed it, and what began as one feature quietly shifted the company's design lens toward motherhood entirely.

Live-design tracker

Most usage happens through the live tracker, not in-app. So I designed it as its own lock-screen surface core functionality parents reach without opening anything.

Designing for the context parents actually use beats in-app depth they don't. A deliberate trade-off that reframed the product the tracker became the primary surface, with logging effort near-zero.

A deliberate break from iOS convention

A Floating Action Button on iOS was a deliberate break from platform convention chosen so mothers multitasking with an infant could reach every primary action one-handed.

Built in FlutterFlow, which gave us the flexibility to keep this experience unified and user-centric across platforms.

A Floating Action Button on iOS was a deliberate break from platform convention chosen so mothers multitasking with an infant could reach every primary action one-handed.

Built in FlutterFlow, which gave us the flexibility to keep this experience unified and user-centric across platforms.

Simplifying data for parents

A baby's whole day, told as a calm, glanceable story clarity over data density. The focus of this was to incorporate similar components and design assets to build familiarity and clarity.

Co-creative workshops

I ran co-creative workshops with stakeholders to align on the MVP scope turning scattered opinions into fast, shared decisions we could build on without backtracking.

Final Outcome

Final Outcome

Final Outcome

What worked well?

In a lean startup team, I learned to ship fast and learn from experiments.

The MVP-to-V3 approach let us gather feedback gradually and adjust with each release. The ownership, autonomy, and impact I earned over that time is what grew me from a junior to a mid-level designer. The problem space was also challenging and helped me learn about

In a lean startup team, I learned to ship fast and learn from experiments.

What would I do differently?

AI wasn't part of our process back then. Today it would let us prototype rapidly, pressure-test our thinking, and catch product risks early the kind we only discovered by shipping.

AI wasn’t part of our process back then. Today it would let us prototype rapidly, pressure-test our thinking, and catch product risks early the kind we only discovered by shipping.

I'd also tie each hypothesis to clear business and proxy metrics from the start defining what success looks like upfront, then tracking the data to prove whether a bet actually worked,.

I’d also tie each hypothesis to clear business and proxy metrics from the start defining what success looks like upfront, then tracking the data to prove whether a bet actually worked,.