DukaanDost AI: A Local Gemma Assistant for Stock, Orders, and Udhaar
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle . What I Built I built DukaanDost AI , a local business assistant for small Indian distributors supplying biscuits, namkeen, chips, and other packaged goods to shops. Consider a simple transaction: a shop orders several cartons, pays part of the bill, and promises to pay the rest later. The distributor needs…
DukaanDost AI is a specialized local business assistant designed for small Indian distributors who supply packaged goods like biscuits, namkeen, and chips to shops. The application simplifies inventory tracking, sale recording, and handling unpaid balances owed to the distributor, known as udhaar. It offers inventory updates, purchase adjustments, low-stock alerts, and generates sales reports, printable receipts, and CSV exports for analysis.
The user-friendly AI assistant can communicate in both English and Hinglish, catering to business inquiries from the distributor. It supports multi-product orders with various payment methods such as cash, UPI, and credit. Additionally, the system provides estimated gross profits and enables CSV export for further record-keeping.
DukaanDost is built using React and TypeScript for the frontend and FastAPI for the backend, with SQLite storing business records. The AI component utilizes Gemma 3, specifically gemma3:1b, running locally through Ollama, which interprets business requests into structured intents and performs validations, product checks, and price calculations. For more complex queries, predefined database queries generate responses, while order creation requires human confirmation.
A challenge faced during development was ambiguous product names, which the app addresses by asking for clarification instead of making assumptions. The project passed 28 out of 30 development checks during a Gemma evaluation, though two outstanding-balance queries were noted for needing further refinement. Despite these limitations, the AI model remains functional when Ollama is unavailable.
The local inference model benefits distributors by eliminating the need for paid inference APIs or proprietary services. Gemma runs on the same computer as the application, and SQLite allows operators to manage their databases independently. The model is flexible, adapting to advancements in local hardware and AI models. The entire application is MIT licensed, and the source code is available on GitHub for review.
Future steps include testing the practicality and willingness of distributors to pay for setup and local support.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
