AI product design · Vibe coded
AI Banking Copilot
From AI concept to working prototype in 6 days.
- Role
- Product Designer
- Duration
- 6 days
- Type
- AI concept + vibe-coded prototype
- Stack
- Figma · Claude Code · Cursor · React Native · Expo · GitHub
- Repository
- Public on GitHub
Try the live prototype
Open in a new tab




01 · Overview
Overview
Designing an AI-native banking experience and transforming it into a working prototype using AI-assisted development.
Instead of building another banking app, this project explores how AI understands context, recommends next-best actions and orchestrates complete workflows.
02 · Why it matters
Banking has always made customers do the thinking.
The AI Copilot changes who does the thinking.
Traditional banking expects customers to search, compare, understand products and decide what to do next.
AI Banking Copilot shifts that cognitive work to the assistant. It understands intent, reasons with customer context and recommends the best next action.
- Traditional banking · Customer does the thinking
- Goal → Search → Compare → Fill forms → Submit
- AI Banking Copilot · Customer states a goal
- Goal → AI understands → AI reasons → AI recommends → Customer confirms
03 · AI workspace
Four workflows, one copilot
Each workflow starts from what the customer is actually trying to do, and ends in a completed action rather than an answer.
Explain transactions

- Answers in plain language first: the merchant and the reason for the charge, before any transaction detail.
- Backs the answer with the record (amount, merchant, date and reference), so the customer can check the explanation for themselves.
- Ends on the three actions that actually follow: view the bill, download the receipt, or report a problem.
- Leaves the thread open, so the next question continues the conversation instead of restarting it.
Analyze spending

- Asks two questions before answering, and no more: which accounts, and which timeframe.
- Leads with the total and its movement against the previous month, then the category split behind it.
- Names the specific things worth knowing, like the category that jumped or the saving that was missed.
- Turns each insight into a recommendation with the control attached: set the budget, view the offers, enable the alerts.
Find best product

- Ranks by net annual benefit against real spending, not by product tier or marketing order.
- Justifies the match with the customer’s own data, such as where the money goes, how often they fly and what they currently earn on it.
- Checks eligibility in place against income, tenure and credit score, without touching the credit file.
- Ends pre-qualified with the terms stated up front: documents, timing, and the odds of approval.
Resolve issues

- Picks the dispute up from the transaction already on screen, with no separate support journey to find.
- Confirms what it understood before acting, and has the customer name the issue type rather than assuming it.
- Files the case itself and shows the summary it submitted, so nothing is raised on the customer’s behalf unseen.
- Returns a ticket reference and a route back to it, so the issue outlives the session.
04 · How I built it
From concept to prototype in 6 days
AI accelerated execution, allowing ideas to move from concept to a working prototype in days instead of weeks.
- Live prototype
- Explore the AI Banking Copilot experience across mobile and web.Try prototype
- Public GitHub
- Inspect the implementation, project structure and development history.View repository
- Built with
- Figma · Claude Code · Cursor · React Native · Expo · TypeScript · GitHub
01
Explore
Benchmark AI experiences
02
Design
Create AI-first workflows
03
Vibe code
Transform into a working prototype
04
Ship
Interactive prototype + GitHub
Conclusion
“The AI does the thinking. The customer stays in control.”
It understands the goal, reasons over the customer’s own data and recommends the next step, but nothing happens until the customer confirms.
Next case studies



