Galla: The Kirana Notebook That Listens
Link to open source: https://github.com/vctc92/galla
Link to Live Project: https://galla-production-8c75d25b.up.railway.app
Galla – Voice-First AI Assistant for Kirana Stores
1. Demo Introduction – Hindi
"मुर्गन अन्ना, दो किलो चावल और एक सर्फ़ खरीदे हैं, हिसाब में लिख दो।"
गल्ला यह सुनकर मुर्गन अन्ना के उधार खाते में ₹230 जोड़ देता है और स्टॉक भी अपडेट कर देता है।
साथ ही बताता है कि सर्फ़ का स्टॉक कम हो रहा है और यह सारी जानकारी हिंदी में बोलकर समझाता है।
2. Demo Introduction – Tamil
"முருகன் அண்ணா ரெண்டு கிலோ அரிசியும் ஒரு சர்ஃப் பாக்கெட்டும் வாங்கியிருக்காரு, கணக்குல எழுதிக்கோ."
இதைக் கேட்டதும், கல்லா முருகன் அண்ணாவோட கடன் கணக்குல ₹230 சேர்த்து, ஸ்டாக்கையும் அப்டேட் பண்ணிடும்.
அதோட, சர்ஃப் ஸ்டாக் குறையுதுன்னு எச்சரிக்கை கொடுத்து, எல்லா விவரங்களையும் தமிழிலேயே சொல்லும்.
3. The Problem
India has over a crore kirana shops, and many shopkeepers still manage customer credit (udhaar) using traditional paper notebooks.
During busy hours, shopkeepers cannot stop to type every transaction. As a result, credit entries get missed, stock records become inaccurate, and collecting pending payments becomes uncomfortable.
Most existing digital ledger apps still expect shopkeepers to type, navigate English menus, and be comfortable using smartphones.
4. Our Solution – Galla
Galla is a voice-first AI assistant designed specifically for local kirana shops.
Instead of learning an app, shopkeepers simply talk to Galla in the language they already speak, including Hinglish, Tanglish, and regional code-mixed languages.
What Galla Can Do
-
Understand local speech: Understands different regional words and variations. For example, arisi, chawal, and चावल all refer to rice. It also understands customer references like "Murugan anna."
-
Perform real actions: Records credit transactions using catalogue prices, deducts stock, checks customer balances, records payments, and initiates supplier reorders.
-
Recover pending payments: Sends customers Telegram payment reminders in their preferred language with UPI payment links and automatically follows up on pending dues.
-
Close the payment loop: When a customer pays, the shopkeeper receives an alert, the ledger updates in real time, and Galla recognizes that the account is cleared.
-
Provide daily business summaries: Shopkeepers can simply ask, "Aaj ka hisaab batao," to hear a summary of credit given, payments recovered, and low-stock items.
5. How We Built It
Our architecture combines real-time conversational AI with an intelligent tool-calling agent.
-
Agora Conversational AI: Handles real-time voice communication and speech recognition.
-
Murf Falcon: Provides natural-sounding Hindi and Tamil voice responses.
-
OpenAI: Powers the intelligent agent, understands user intent, and executes actions through custom tool calls.
-
Custom Catalogue and Ledger: Handles regional product names, customer identification, pricing, credit records, and inventory.
-
FastAPI and SQLite: Power the backend and data management.
-
Telegram Bot API: Sends payment reminders and follow-up notifications.
-
React: Powers the phone application and live dashboard.
6. Complete Demo Journey
Our demo demonstrates the complete end-to-end journey:
-
The shopkeeper speaks naturally in an Indian language.
-
Galla understands regional and code-mixed speech.
-
The AI agent identifies the customer, products, quantities, and transaction intent.
-
It invokes the appropriate tools to update the ledger and inventory.
-
It confirms the transaction verbally in the shopkeeper's language.
-
It sends payment reminders and tracks customer payments through external services.
-
The shopkeeper can ask for a daily business summary at any time.
7. Closing Statement
With Galla, shopkeepers don't need to learn another app, type during a busy sale, or struggle with English menus.
They simply talk the way they already do.
Galla understands their language, manages their business records, and helps them get their money back.
Our goal is simple: make digital business management accessible to every kirana shopkeeper in India, starting with just their voice.
This build was uploaded as a hackathon project


