Design Lead and sole designer · 2022–2024 · Remote

Tapioca

Tapioca was a DeFi app for borrowing and lending across many blockchains at once, with its own stablecoin, USDO. I was its only designer for two years, from the first user flows to the launched product.

Below: the Mint page, with every open position and its loan health.

500+
Screens, every flow designed in full
~150
Components, 3,500 variants
6
Sections, from minting to locking
$8.39M
Raised in the July 2024 token launch

01 · The problem

One action, many signatures

Because Tapioca worked across chains, a single action such as “deposit and borrow” could need several approvals and transactions in the user’s wallet, sometimes with a change of network in between. Each one was a point where people could get lost or give up.

I mapped the logic with the lead front-end developer: which transactions could be sent together, what had to finish before the next could start, and every status a transaction could be in.

Beta testing confirmed it was the biggest problem in the product. When I sorted the feedback by impact and effort, approvals came top: there were too many, people did not understand the steps, and nothing told them a step had worked.

Flowchart of the transaction approval logic, from pressing the action button through batches of wallet approvals to success, with branches for declined and failed transactions.
The approval logic, worked out with the developer before any interface was drawn.

02 · The answer

One panel for every action

It combines steps wherever it can, says in plain language what is happening, numbers the steps, and has a designed state for pending, in progress, failed, rejected and complete.

State

Recreated in HTML and CSS from the Figma components, not the shipped code. Use the arrow keys to move between states.

Transaction pendingTransaction in progressTransaction failedTransaction rejectedTransaction complete

You’re depositingYou deposited

1.0752 ETH$2124.59 on Ethereum

You’re borrowingYou borrowed

250.0123 USDO$250.00 on Arbitrum

USDO

Fees: $1.34

Failed:

Completed:

15:09 UTC, 15 March, 2023LayerZero scan

Approve token for spending

Estimated time:Time remaining:--0:29

There was an error with this transaction

You rejected the transaction

Your transaction is complete!

  1. 1
  2. 2
  3. 3
Please confirm in wallet
Confirming…
CloseTry again?
CloseView positions

I then drew that panel for every action in the product, 145 screens in all, so the developers had a complete record and no state was left to guesswork.

Zoomed-out view of a Figma file containing dozens of rows of small screens, grouped under coloured headings.
One of eight product files. This one holds every transaction flow, laid out for the developers.

03 · The system behind it

Parts no library provides

We adapted Material UI so that design and code shared the same components. On top of it I built the parts a multichain lending app needs and no library provides.

Every screen worked in dark and light mode from the same components.

LTV Ratio12 variants
Loan to Value Ratio slider at 20%, green, with quick-pick buttons for 25%, 50%, 70% and Max.Loan to Value Ratio slider at 20%, green, with quick-pick buttons for 25%, 50%, 70% and Max.The same slider at a moderate level, amber.The same slider at a moderate level, amber.The same slider at a risky level, red.The same slider at a risky level, red.
Deposit boxInput + token
Deposit field showing 54 sAVAX on Avalanche, worth $1000.00, with the wallet balance beneath.Deposit field showing 54 sAVAX on Avalanche, worth $1000.00, with the wallet balance beneath.
Dark mode toggle4
Click it
StepperIn the panel
Three numbered steps, the first active.Three numbered steps, the first active.First step ticked, second active.First step ticked, second active.All three steps ticked.All three steps ticked.
Token on network · Token select menu116 variants
ETH on Ethereum.ETH on Ethereum.ETH on Optimism.ETH on Optimism.ETH on Polygon.ETH on Polygon.USDO on Arbitrum.USDO on Arbitrum.USDO on Avalanche.USDO on Avalanche.BNB on BSC.BNB on BSC.

Try the token picker

Every token on every supported network, with a disabled state for each. The token always shows which network an asset is on, because the same token on two networks is not the same thing to the user. Balances in the picker are Figma’s sample values.

Liquidation risk · Loan states10 + 20
Zero risk chip.Zero risk chip.Low risk chip.Low risk chip.Medium risk chip.Medium risk chip.High risk chip.High risk chip.Max risk chip.Max risk chip.Low risk label.Low risk label.Moderate risk label.Moderate risk label.High risk label.High risk label.
Alert text descriptions24 variants
Alert warning that the position is close to insolvency.Alert warning that the position is close to insolvency.Alert about gas.Alert about gas.Alert that USDO is above its peg.Alert that USDO is above its peg.
Network tag18 variants
Ethereum mainnet tag.Avalanche tag.Optimism tag.Polygon tag.Fantom tag.BSC tag.Arbitrum tag.Moonriver tag.

04 · Three smaller decisions

Details that carried weight

Loan health at a glance

A note from an early team workshop asked for a “battery / visual way of showing health”. It became a colour-coded bar on every position, so the risk of a loan is readable without knowing what loan-to-value means.

Three open positions, each with a green, amber or red loan-to-value bar.Three open positions, each with a green, amber or red loan-to-value bar.

Saying what you can lend

Beta testers expected to lend the assets they held, such as ETH, when the protocol only lent USDO. The Lend page says so in its title.

Rewards that expire

Locked positions earned an option each week that had to be used before it ran out. The table shows what is available, used or waiting for each lock, with a countdown.

The Lock page: a table of locked positions showing the reward available for each, its price and when the lock expires.

05 · What happened

Launch, and after

Tapioca launched in October 2024, after raising $8.39M in its July 2024 LBP (liquidity bootstrapping pool). On 18 October it lost about $4.4M in a social-engineering attack, and it later shut down. A shame. 💀

I wrote more about the design process on Medium.