Taking an AI Chief of Staff from a broken pre-MVP to a funded, launched product — as the sole designer.
TheTop connects to your inbox, calendar and more, and every morning synthesizes one decision-ready brief — priorities, key events, and the signals that need attention (commitments, decisions, opportunities, risks). A decision layer, not another chatbot: it does the work before you ask.
I joined at a broken pre-MVP stage and, as the only designer, took it to a live, launched product with real users. Outcomes:
The real design problem was never the interface. It was trust — and every decision below is about earning it.
I owned it end-to-end — research → concept → UI (web + mobile) → design system → marketing site → dev handoff — working directly with the founders and dev team.
I inherited a design and a partial build that wasn't launchable. Worse, it behaved like a filterable dashboard — it surfaced everything and made the user do the sorting. For a product whose whole promise is push ("it already did the thinking"), that was the wrong model.
And the harder constraint underneath: TheTop needs deep access to your inbox and calendar to work at all — but its audience is founders, investors and sales leaders, skeptical people who won't grant that access until the product has earned it, and who drop the moment something smells like generic AI.
So the real job wasn't "polish the UI." It was: define what the MVP actually was, make it shippable fast, and design the whole thing to earn trust before it asks for access.
Early on I audited the inherited flow against the product's own positioning and against tools this audience already trusts — Superhuman, Linear, Notion, Granola, Sunsama. Two things were clear: the flow contradicted the push positioning, and it was adding surface (filters, feeds) where those tools win by removing it. Later, once analytics were connected, the data pointed to the exact leaks behind two of the decisions below.
Flipped the product from something you operate into something that delivers. Kept web, but made mobile the primary surface — an assistant in your pocket — opening on a single glanceable morning synthesis instead of a feed you filter.
Result: Mornings became the product's core loop — users open to a brief, not a to-do list.
Analytics showed users dropping exactly at the connect-email step — an order problem, not a copy one. I moved sign-up and email access to the end, after a labeled example brief, and cut the flow 17 → 12 screens. You don't argue a skeptic into trust; you let the product earn it.
Result: Onboarding completion rose from 68% to 84%, with drop-off at the email step nearly eliminated.
A free tier kept users comfortable without converting. I removed it and instead let users experience a real brief — the product's core value — before the paywall, then moved them onto a trial (7-day with card / 3-day without).
Result: +80% paid conversion — users now hit the paywall after seeing real value, not before.
Making a complex supplement routine simple enough that people actually keep it.
Vitamin Book helps people track their daily intake of vitamins, minerals and supplements — add items, understand what's in them, set reminders, and never miss a dose. I designed the multi-screen experience — from the vitamin library to scheduling and history. Shipped and live on the App Store.
UI/UX at Martspec (Aug 2023 – Aug 2024, across several products). On this project I designed the Vitamin/Health Tracker's mobile experience — the vitamin & mineral library, nutrient detail, scheduling & reminders, history/trends, and subscription. Worked directly with the engineering team.
Supplement tracking has two hard problems at once. It's complex — people take many different vitamins and minerals, each with its own dosage and timing — and adherence is low: users start a routine, forget a dose, and quit. A tracker only works if people come back every day, so the real design goal wasn't more features — it was making a complex routine feel light enough to keep.
Organized everything into categorized lists (vitamins, minerals, pills) with detailed nutrient cards, so users can find an item, understand what it does and its dosage, and add it without friction. Structure first — you can't keep a routine you can't see.
The Schedule section lets users set dosages, choose dates and get reminders — the core loop against "I forgot." Logging a dose is fast; the app does the remembering so the user doesn't have to.
A history view — a calendar of taken/missed doses and trend charts — turns invisible daily effort into something you can see and feel good about. Visible progress is what turns a tool into a habit.
Plus a nutrient/ingredient breakdown and food analysis for informed intake, and a clean subscription flow.
"Julia's ability to create user-friendly interfaces for both web and mobile is truly impressive… her design contributions to the Vitamin Book app showcase her ability to design intuitive and engaging multi-screen experiences."
— Alex Protyagov, Lead Software Engineer, MartspecReflection. Designing a health tracker taught me that the hard part isn't the feature list — it's adherence. An app like this only works if people return daily, so every decision optimized for making a complex routine feel light: fewer taps to log, reminders that do the remembering, and a history that rewards consistency. The interface's job was to make staying on top of your health feel easy, not clinical.
A focused redesign to turn browsers into buyers.
SnapMyTCG is a live TCG card scanner — scan a card, see its live market value, and track your collection as a portfolio. I was brought in as the solo designer to redesign the onboarding and paywall and lift monetization.
Result: increased subscription conversion — a higher share of users who reached the paywall subscribed. Shipped and live in the App Store.
Product designer, ~1 month (May 2026). I owned the redesign of onboarding, the paywall/monetization flow, and UX improvements across the core screens, working directly with the client's development team. No other designer.
The app worked — scanning, pricing and the portfolio were all there — but the business didn't: subscription conversion was low. Users installed, scanned a few cards, and never subscribed. Two gaps behind it:
My brief: redesign onboarding and the paywall to convert more of them.
Rebuilt the flow to show the magic first — scan → live price, the portfolio, "find new cards" — backed by social proof, so users understand why it's worth paying before they're ever asked. Value first, commitment second.
Redesigned monetization as layered surfaces instead of a single wall — each meeting the user at a different moment of intent:
Tightened the screens users hit most — card detail (card → live price → Add to Portfolio in one glance) and the portfolio value view — so the path from scan → worth → upgrade stays obvious.
Reflection. This project sharpened how I think about monetization: a paywall doesn't convert on its own — onboarding has to make the value obvious first, so that by the time users reach the paywall they already want what's behind it. The biggest lever wasn't the paywall screen itself, it was the sequence — show the value, then ask.
An AI stylist is only as good as what it knows about you — so the design job was to make users teach it their taste, effortlessly.
Muse merges style inspiration with your real closet: share looks and follow your circle's fits, digitize your wardrobe, and get outfits from an AI stylist that knows what you own. Shipped to the App Store — 82% of new users completed the style quiz, giving the AI a taste profile before their first session.
Product designer, on an existing app. I owned the design of the onboarding taste-quiz, the style feed, profiles, the wardrobe, and the AI-stylist experience. No other designer.
An AI stylist has a cold-start problem: it can't style you until it knows your taste and what you own — but asking users to input all that upfront is exactly where they drop.
The design challenge: get people to invest that information before they've gotten any value, without it feeling like work.
Instead of dropdowns, onboarding is visual: tap the looks you'd actually wear, pick the brands that feel like you. It reads as fun, not setup — and it quietly builds the taste profile the AI needs.
Result: 82% completion, so most users reach their first session already understood.
The wardrobe turns a user's actual clothes into a structured, browsable grid (tops / bottoms / shoes / accessories) — the raw material that lets the AI build outfits from what you own, not generic catalog looks.
A FRIENDS / FOR YOU feed of real outfits, follows and saved collections gives Muse a reason to open daily — inspiration from your circle, not just a chatbot.
(The payoff — Ask Muse — ties it together: an AI stylist that answers "what works together?" and builds a palette from your own pieces.)
Reflection. Muse taught me that with AI products, onboarding isn't a step before the product — it is the product's intelligence. The model can only be as personal as what users are willing to teach it, so the job is to make giving up that data feel like play, not setup. Getting 82% through the quiz mattered more than any screen after it — a stylist that doesn't know your taste has nothing to say.
Think deeply about people, products, and the future of AI interactions. For the past years I've gone deep on product design for AI — turning fuzzy, agentic ideas into clear, trustworthy interfaces — and on the systems behind them: design systems, research, prototyping, and motion. Lately I'm having the most fun taking a design past Figma and building it myself.
Shots, explorations, and UI ideas that didn't need a full case study — just a good look. More on Dribbble.