HishabBondhu — Voice-Powered Digital Khata
The Problem
Many small grocery (mudi) shop owners in Bangladesh still manage their daily sales and customer credit (baki) using paper notebooks. For shopkeepers who may not be comfortable with complicated digital applications, keeping track of cash sales, credit sales, customer dues, and daily transactions can be difficult.
A customer may take ৳5,000 worth of products on credit, while another customer may make a small ৳50 purchase without even giving their name. Managing all of these transactions manually can lead to mistakes, forgotten dues, and inaccurate records.
Our Solution
HishabBondhu is a Bangla-first, voice-powered digital khata designed specifically for small grocery shop owners.
Instead of filling out complicated forms, a shopkeeper can simply speak naturally.
For example:
“রিফাত ১০০০ টাকার চাল বাকিতে নিয়েছে।”
HishabBondhu understands the statement, identifies the customer, amount, product, and transaction type, asks for confirmation, and updates the digital ledger.
It also supports anonymous customers:
“৫০ টাকার হলুদ বিক্রি করলাম।”
No customer name is required for this type of sale.
How It Works
The core workflow is:
Voice → Speech-to-Text → Gemini AI → Structured Data → Validation → Confirmation → Database → Updated Ledger
Gemini is used to understand natural Bangla and Bangla-English mixed speech and convert it into structured transaction information.
For example:
{
"transactionType": "DUE_SALE",
"customerName": "রিফাত",
"product": "চাল",
"amount": 1000
}
The AI does not directly modify financial balances. After the AI extracts the user's intent, the application validates the information, asks for confirmation, and then uses deterministic business logic to update the database.
This keeps the AI responsible for understanding language, while the application remains responsible for financial calculations and data integrity.
Key Features
🎤 Voice-Based Transactions
Shopkeepers can record transactions simply by speaking.
Examples:
- “রিফাত ৩০০ টাকার মাল নগদে কিনেছে।”
- “সিফা ৫০০০ টাকার মাল বাকিতে নিয়েছে।”
- “২০০ টাকার লবণ বিক্রি করলাম।”
- “রিফাত ৫০০ টাকা বাকি দিয়েছে।”
👤 Known and Unknown Customers
HishabBondhu does not force every transaction to have a customer name.
Known customers can have their own digital ledger, while anonymous walk-in sales can be recorded without creating a customer.
New customers can also be created dynamically through voice.
📒 Digital Baki Khata
Each known customer can have:
- Current outstanding balance
- Total credit taken
- Total amount paid
- Transaction history
- Last transaction
🧠 Ask Your Khata
Shopkeepers can ask questions naturally in Bangla, such as:
“আজকে মোট কত টাকা বিক্রি হয়েছে?”
“রিফাতের বাকি কত?”
“কার কাছে সবচেয়ে বেশি টাকা বাকি?”
“এই মাসে মোট বিক্রি কত?”
HishabBondhu retrieves the actual information from the database and responds in simple Bangla.
🔊 Bangla Voice Responses
The system can respond with short, clear Bangla answers so that shopkeepers do not always need to read information from the screen.
↩️ Safe Transaction Handling
Before changing financial records, the system shows a confirmation.
It also supports transaction reversal, manual correction, and clarification when the AI is unsure about important information.
Why HishabBondhu?
We are not trying to build another complicated accounting application.
We are building a system around the way a small shopkeeper actually works.
A shopkeeper should not have to learn accounting software just to record a ৳50 sale or remember who owes them ৳5,000.
With HishabBondhu:
Just speak. The হিসাব is recorded.
Challenges
One of our biggest challenges was understanding natural Bangla speech because people do not always use the same sentence structure.
For example, a shopkeeper might say:
“রিফাত ৫০০ টাকা বাকি নিয়া গেছে।”
or:
“রিফাতের নামে ৫০০ টাকা বাকি রাখো।”
Both sentences should produce the same type of transaction.
Another major challenge was ensuring that AI could not accidentally create incorrect financial balances. We solved this by separating AI-based language understanding from deterministic financial calculations.
What We Learned
This project taught us that building a useful AI application is not just about connecting an LLM to a user interface.
For a real-world financial application, AI needs to work together with:
- Authentication
- Database design
- Validation
- Structured data
- Financial business logic
- Confirmation workflows
- Error handling
- User-centered design
Most importantly, we learned that AI becomes much more useful when it is designed around the real needs and habits of its users.
Future Plans
We plan to expand HishabBondhu with:
- Automated due reminders
- SMS and WhatsApp notifications
- Inventory management
- Supplier management
- Profit estimation
- Product-level analytics
- Offline-first functionality
- Multi-user shop accounts
- Advanced voice-based business reports
Our Vision
HishabBondhu aims to make digital bookkeeping accessible to small grocery shop owners who may not be comfortable with traditional accounting software.
Instead of forcing shopkeepers to learn complicated technology, we want technology to understand the way they naturally communicate.
Speak naturally. Keep the হিসাব. Know your business.
Log in or sign up for Devpost to join the conversation.