I Built an AI Expense Tracker That Understands the Way People Actually Type

Most expense tracking apps expect users to type in a fixed format.

But real people don’t track expenses like that.

Some people type:

“Paid 100 for petrol”

Some type:

“500 रुपये का पेट्रोल भराया”

Others naturally mix languages:

“Petrol ke liye 300 pay kiya”

And sometimes people combine everything together in a single sentence.

Most expense trackers completely fail at understanding this.

That’s the problem I wanted to solve while building Vitmora.

The Problem With Traditional Expense Tracking

I’ve tried many expense tracking apps over the years.

Most of them were good at charts, dashboards, and reports.

But almost all of them had the same problem:

They expected users to adapt to the app.

You usually have to:

Select category
Enter amount
Choose payment method
Add notes
Fill forms repeatedly

After a few days, it starts feeling like work.

And honestly, most people stop tracking because of that friction.

But there was another problem I kept noticing.

Language.

People Don’t Think About Money in One Language

In countries like India especially, people naturally switch between languages while typing.

Sometimes it’s English.

Sometimes Hindi.

Sometimes Hinglish.

Sometimes regional languages.

And most of the time, it’s mixed naturally.

That’s how people actually communicate.

For example:

“Paid 100 for petrol”
“500 रुपये का पेट्रोल भराया”
“Petrol ke liye 300 pay kiya”
“Swiggy order 450 paid by UPI kal night”

These all mean something very clear to humans.

But most apps struggle to understand them.

That’s where the idea behind Vitmora started.

The Goal Was Simple

I didn’t want users to learn how to use the app.

I wanted the app to understand users naturally.

Instead of structured forms, the experience should feel more like chatting.

Just type normally.

Vitmora should figure out:

amount
category
payment method
merchant/context
intent

automatically.

Building Multilingual Expense Understanding

This turned out to be much harder than it sounds.

Because users don’t type consistently.

For example:

“Petrol ke liye 300 diye”

“Petrol bharaya 500”

“500 ka fuel”

“Bike me petrol 200”

All of these technically describe the same kind of expense.

Then there’s mixed language input.

A single sentence might contain:

English words
Hindi words
regional language
abbreviations
spelling mistakes

And users still expect the app to understand them instantly.

That became one of the biggest product and engineering challenges while building Vitmora.

Understanding Intent Instead of Perfect Grammar

One thing I realized very quickly:

People don’t type for machines.

They type for convenience.

Nobody wants to stop and think:
“Am I formatting this correctly?”

The app should understand intent even if the sentence isn’t perfect.

For example:

User Input	What Vitmora Understands
Paid 100 for petrol	Fuel expense
500 रुपये का पेट्रोल भराया	Fuel expense
Uber ke liye 300 pay kiya	Travel expense
Swiggy order 450 paid by UPI	Food delivery + UPI
Coffee 120 with card	Food & card payment