Case Study

Asaan Bill

A voice-powered invoicing app that lets Pakistani shopkeepers create and share invoices in under 60 seconds — without typing a single word.

GitHub

FrameworkFlutter (Dart)

State ManagementRiverpod 2

Local DatabaseIsar Community

NavigationGoRouter

Speechspeech_to_text (Google)

Sharingshare_plus · url_launcher

Background

The Problem

Tariq runs a busy general store in Lahore's Raja Bazaar. On a good day he serves 80–100 customers. For years, his invoicing system was a stack of paper receipts, a ballpoint pen, and a prayer that his handwriting was legible enough for the customer to read.

He tried switching to a basic spreadsheet on his phone. It took him 10–15 minutes per invoice — too slow when there's a queue of five people waiting. He had to type every item name in English, manually calculate totals, and screenshot the screen to send over WhatsApp. He went back to paper within a week.

The real cost wasn't just time — it was the professional image he was losing every time he handed a customer a crumpled handwritten chit instead of a clean digital invoice. What Tariq needed wasn't a complicated accounting app. He needed something he could talk to in Urdu, that would handle the math, and send the invoice to the customer's WhatsApp in two taps.

"He needed something he could talk to in Urdu, that would handle the math, and send the invoice to the customer's WhatsApp in two taps."

Solution

Key Features

Voice-First Item Entry

Say the item name in English, then the quantity and price in Urdu. The app converts spoken Urdu number words — "panch sau", "do hazar" — into digits automatically. Adding a full item takes 5–8 seconds without touching the keyboard.

Session-Based Invoice Building

Add multiple items in one continuous voice session. Each saved item appears in a running list so the shopkeeper can see everything added so far, edit any item heard incorrectly, and keep going without closing anything.

WhatsApp Direct Sharing

One tap opens the customer's WhatsApp chat directly with the invoice pre-loaded as a message. No copy-pasting, no manually finding the contact.

Fully Offline

Invoices are saved to the device instantly using Isar — a fully offline local database. If the internet drops mid-session, nothing is lost and nothing slows down.

Engineering

Key Challenges

Voice recognition breaking on MIUI devices after the first session

The Problem

On Xiaomi phones running MIUI — the most common low-cost Android in Pakistan — the speech engine stops firing the "done listening" callback after the first session closes. The mic button would appear stuck on "Listening..." indefinitely, breaking the core feature for the majority of target users.

The Solution

A 300ms hardware release delay between sessions. A reinitialize flag that triggers a fresh engine setup when a permanent error is detected. A fallback timer that force-resets the UI if the platform goes silent. Switching the mic button from onTap to onTapDown to bypass MIUI's aggressive touch rejection.

Recognizing mixed Urdu-English speech for numbers

The Problem

Pakistani shopkeepers don't speak in one language. A shopkeeper might say "cooking oil" (English) then "teen kilo" (Urdu) then "chaar sau pachas" (Urdu). The English speech engine simply doesn't know what "teen" or "panch sau" means.

The Solution

The app switches the speech locale dynamically — English for item name fields, Urdu (ur_PK) for quantity and price fields. A custom number conversion function maps over 80 Urdu words, romanized transliterations, and compound phrases into digits before saving. Partial speech results are used to accept a recognized number the instant it's detected, reducing wait time from 3–4 seconds to under 1 second.

Screens