I Built a Voice-First AI Khata for a Shopkeeper Who Shouldn’t Have to Type
This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend What I Built I built Khata Ledger , a voice-first AI credit ledger for Family Friend , who runs Baba gernal store in Ludhiana . Small neighbourhood shops often give groceries and other essentials on credit. These transactions are usually written in a notebook, saved in scattered messages, or simply remembered. That…
This is a story about a shopkeeper in Ludhiana who needed a way to manage on-credit transactions without having to type everything down. Many small neighborhood shops still rely on notebooks, memory or scattered messages to keep track of customer payments and dues. When there are many customers and transactions, it becomes difficult to maintain accurate records.
Amrit Kang built Khata Ledger, a voice-first AI credit ledger that allows shopkeepers to speak their transactions instead of typing them. The app uses ElevenLabs' transcription service to convert speech into text and then extracts key details like customer name, amount, transaction type and due date. This information is then safely added to the shop's private ledger.
The app has several key features. Customers can be created and managed as private accounts. The shopkeeper can record voice notes directly in the browser and the app converts these into structured ledger entries. It converts Hinglish and Hindi spoken by the shopkeeper into JSON data. If the AI model is uncertain about a transaction, it is flagged for manual review by the shopkeeper rather than automatically being added to the ledger.
The app also reminds the shopkeeper of payment due dates, generates reports of a customer's payment history, and provides a passbook view of all transactions. The project uses modern technologies like FastAPI, SQLite, Gemma 4B, Ollama, ElevenLabs Scribe v2, MongoDB Atlas and more. The complete source code is available on GitHub with a demo live application.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.