A working prototype of a goal-based savings app with an AI coach. It reads your accounts with consent, shows where each goal really stands, and suggests changes it can explain. It never moves money.

TypeAI Experiment
ToolsClaude
PlatformMobile app prototype
Year2026
THE QUESTION

Can an AI coach help people save for goals they care about, without becoming one more app that guesses, nags or sells?

MY ROLE

Problem framing, the system’s rules, UX and visual design, motion, and directing AI to build and test the prototype.

THE APPROACH

Write the rules first (what the app may see, say and change), build every scenario on real numbers, then audit it the way a hiring panel would.

THE RESULT

A clickable prototype with five scenarios and an AI coach that answers from your own data. I also found and triaged 78 dead ends before anyone else saw them.

In this case study
01

Most people know what they want to save for. The plan falls apart in month three, when life does something the spreadsheet didn’t expect.

Budgeting apps track what already happened. They rarely say what a good month means for a goal, or what a bad one should change. I wanted to see whether an AI layer on top of a person’s real accounts could turn each month into one clear, explainable next step, while every decision stays with the person.

Three goals share one small surplus, so a single good or bad month moves all of them.
02

I wrote the rules before any screen: what the app can see, what it can say, and what it must never do.

Account data comes read-only through India’s Account Aggregator framework, with a statement upload for co-op banks that aren’t on it yet. Goal money is a label inside the person’s own bank, never a transfer. The coach answers only from linked data, shows its sources, declines stock tips and changes nothing until the person applies a plan. Then I mapped one month across six moments, and where a plan usually breaks in each.

Service blueprint · one month with Wealth Flow

One salary, six moments, one plan

{{p.num}}{{p.name}}

{{r.who}}{{r.what}}
0
rupees the app can move on its own
5
scenarios, from a first open to a salary that drops
78
dead ends found in my own audit, then triaged
Linking and the first goal earn permission. The app only speaks up at the moments after that.
03

I worked with Claude as a design partner and builder. I set the rules and the bar, it drafted, and I reviewed every screen, then tried to break it.

The call

The coach never moves money.

Why

Trust is the product. The coach runs what-ifs on your real numbers and proposes a plan, but nothing changes until you apply it, and every change can be undone. When a question needs a SEBI-registered adviser, it says so Guidance, not advice

The trade-off

It can’t act for you or pick investments, so some answers end with what it can show you instead of a tip.

From there I worked in short loops on each screen: a calm dark system with one accent kept for AI and progress, motion only where it explains a change (a tick that draws, a ring that fills to what’s saved, a chart that labels every month), and copy that states facts instead of praise.

The insight · a what-if · where it stops

Every suggestion shows why

The coach explains, compares and stops at its limits. The person decides.
04

A working prototype across five scenarios, from linking accounts to a salary that drops. Then I reviewed it the way a hiring panel would.

That audit found 78 dead ends: numbers that disagreed between screens, links that went nowhere, and states the app could get stuck in. I fixed the 25 that mattered most, in four batches, and wrote a one-line defence for each of the rest.

Home · salary drops · a new plan

When life changes, the plan follows

The app asks before it changes anything, then shows the new dates for every goal.
Working prototype

Try Wealth Flow yourself

Link accounts, set a first goal, act on an insight, ask the coach a what-if and see what happens when a salary drops.

5 scenarios · about 4 minutes
05

  • Writing the rules first, what the app may see, say and change, made every screen easier. Most of the hard decisions had already been made.
  • An AI coach earns trust through what it refuses to do and how easily it can be undone, not through how much it knows.
  • Reviewing my own prototype like a hiring panel found more than any single design pass. In a money app, the same number has to agree on every screen.

Next Project

SEE ALL WORKS
Cleared app on a phone, with a notification reading Cleared to go home, insurer approved in 2 hours 45 minutes Cleared
ClearedAI EXPERIMENT
Let's talk
akhilbhargavan70@gmail.com
© 2026 Akhil Bhargavan
{{toastMsg}}