← All AI work

From a customer problem to production Android XML

Every feature now gets a working AI prototype the team can try before development, and ships as Android XML on our design system.

My role
Design and prototype; the developer wires the data
Built with
Claude · Codex · Claude Design · Figma Make
Example
Split Payment
01 · Research

Pixel-perfect first, edge cases later

Every feature started as a pixel-perfect Figma design, then went for approval and stakeholder talks. When we changed an existing feature or added a new one, edge cases were often missed.

02 · Ideation

Read the code first, then let people try it

AI reads the live code before any design work. A quick prototype then lets the team try the idea before anyone polishes pixels.

03 · Solution

Five steps from problem to production code

A PRD, customer insights and Amplitude go in. Production Android XML comes out, and results after launch feed the next round.

My decisions

The orange diamonds. AI never passes a gate on its own.

Design system

Feeds both the prototype and the code, so they match.

Dotted lines

A "No" at a gate loops back.

Example: Split Payment, which lets a customer pay one bill in more than one way. Today's screen is rebuilt first, then the new feature is built on top.

The live Zobaze app counter screen
Live app
The same counter screen rebuilt in Claude Design
Rebuilt in Claude Design
Receipts list where split bills show as by Cash and Debit Card
New: split bills on receipts
04 · Success

Everyone sees the same thing, sooner

Edge casesCaught before design, not after the build
FasterStakeholders align on the problem and the fix

Want to see this workflow live?

I'm happy to walk through a feature from user signal to shipped screen.

vivektravi@gmail.com LinkedIn