← All work

Case study 01 · Fintech UX · Accessibility

roomie
receipts

Splitting shared bills without the awkward conversation.

Role
Solo UI/UX Designer
& Frontend Developer
Timeline
v1 Nov 2025 – Feb 2026
v2 Nov 2026
Tools
Figma · React (Vite)
Supabase · Cloudflare Pages
Live
roomie-receipts.pages.dev ↗ (opens in a new tab)
Roomie Receipts v2 landing page on desktop

The
problem

Living with roommates meant constant small shared costs: grocery runs, furniture for the house, trips together. Keeping track of who paid for what was harder than it should be, and asking a roommate to pay me back felt even more awkward.

Existing tools like Splitwise required everyone to download an app and create an account. For a quick grocery split, that setup felt like more effort than the problem deserved.

Goal

Split a bill and share it with roommates in under a minute. No app download, no sign-up.

v1: What I shipped

Used by my roommates and friends
for groceries, furniture and trips.

No login
Create a receipt, add names, and get a share link immediately.
Two split modes
Split equally, or enter custom amounts. With two people, one amount fills the other.
Share flow
Copy the link, or a ready-made friendly reminder for the group chat.
Retro style
Bungee type and bold outlines to make money talk feel lighter.
v1 create receipt screen on desktop
v1 split screen on desktop

Auditing v1,
one year later

After learning WCAG in class, I reviewed v1's code and design with fresh eyes. It worked for my friends, but not for everyone.

  1. Accessibility

    • Body text in an ultra-thin, all-caps display face
    • Contrast as low as 1.8:1 (AA needs 4.5:1)
    • Inputs with placeholders but no labels
    • Checkboxes and history items unreachable by keyboard
    • Errors only as browser alert() pop-ups
  2. Money accuracy

    • $10.00 split three ways gave $3.33 each: one cent vanished
    • Custom splits could save without adding up
    • Totals were never saved: a missing column, and the error was ignored
  3. Privacy & trust

    • The README promised local-only privacy; receipts actually lived in Supabase
    • Row Level Security was off, so the public key could read every row
    • Advertised receipt scanning and PDF export that didn't exist
  4. Device support

    • Designed and tested only on laptops
    • Only the landing page adapted to small screens, though people split bills on their phones

Four decisions

each keeps v1's personality
while fixing what the audit found

  1. Keep the fun, fix the legibility

    Options
    Drop the retro style · keep it everywhere · display text only
    Shipped
    Bungee only for headings and buttons; Space Grotesk for body, inputs and amounts; no text-stroke.
    Why
    Every text colour now passes 4.5:1. The receipt title went from 2.1:1 to 5.9:1.
    v1 landing page on a phone: thin all-caps text with low contrastv2 landing page on a phone: readable body font and stronger contrast
  2. Make every cent visible

    Options
    Round silently · give the cent at random · show it and let people choose
    Shipped
    Integer-cent math. “1¢ left over: who covers it?” with a dropdown, and a live “left to assign” counter.
    Why
    In money UX, trust comes from totals that always add up. The server re-checks the sum too.
    v2 split screen asking who covers the 1 cent left over
  3. Be honest about where data lives

    Options
    Add accounts · secure the database · store everything locally
    Shipped
    “The link is the key”: three single-receipt Postgres functions, RLS on, and a plain privacy note.
    Why
    No-login was the core value, so the fix belongs in the system and the copy, not in more friction.
    v2 share screen with a privacy note explaining that anyone with the link can view the receipt
  4. Make it work on phones

    Options
    Shrink the desktop layout · build a separate app · responsive layouts
    Shipped
    Below 768px: stacked sections, full-width buttons, 44px tap targets, 16px inputs, number keypads.
    Why
    Splitting usually happens right after paying, phone in hand.
    v1 create receipt form squeezed onto a phone screenv2 create receipt form with stacked full-width fields on a phone

Outcome

v2 shipped Nov 9, 2026,
live at the same link.

Lighthouse accessibility score
Screenv1v2
Landing88100
Create receipt96100
History96100

The scores understate the change. Automated checks can't tell that v1's checkboxes were unreachable by keyboard, that its hairline text was hard to read, or that its numbers didn't add up. Those were found by reading the code and testing by hand.

What testers said

Roommates & friends, real splits

  • No login was the favourite part: they could start right away.
  • The copy-a-friendly-reminder button stood out. Nobody had to write the “you owe me” message themselves.
  • They asked for item-level splits for groceries, and named line items (gas, Uber, desk, sofa) for trips and furniture.

What's next · v3

From one total to a list of items

Both requests point the same way: named items, each split among the people who shared it. v2's cent-accurate math is the foundation, since each item is split in integer cents before the totals are added up.

what a year
changed

v1 taught me to ship fast; revisiting it taught me what fast had skipped. A design can feel delightful to its maker and still exclude people who rely on keyboards, screen readers or good contrast. And when an app handles money, “close enough” rounding and vague privacy claims quietly erode trust. Next time, accessibility and data honesty are part of the first draft, not a later fix.