[Entry // 2026.10 · 2 min]

I Built Dukaandaar AI: An Open-Source AI Assistant for Small Businesses

For Hacktoberfest's “Build for a Friend” challenge, I built Dukaandaar AI for my friend, who works as an accountant handling inventory-related operations.

Hacktoberfest

Most projects start with an idea. Dukaandaar AI started with a person.

For Hacktoberfest's “Build for a Friend” challenge, I built Dukaandaar AI for my friend Suman Mandal, who works as an accountant handling inventory-related operations.

Seeing how much small-business work involves managing inventory, sales, purchases, expenses, and reports made me wonder:

What if you could simply talk to your business instead of navigating through it?

What is Dukaandaar AI?

Dukaandaar is an open-source AI-powered business management application for small businesses.

It combines traditional business tools with an AI interface for: 📦 Inventory 🛒 Sales & purchases 💰 Expenses 📊 Reports 🤖 Natural-language queries 🎙️ Voice interaction

The goal isn't to put a chatbot next to a business application. I want AI to become an actual interface for interacting with business data. For example:

“I received 20 units of Maggi from Raj Traders at ₹12 each.” The system should understand the intent, show what it plans to change, ask for confirmation, and then update the relevant records.

Privacy First Business data can be sensitive, so I didn't want Dukaandaar to depend entirely on third-party AI services. The project supports two approaches: 🏠 Local / Private — run the application and AI locally, keeping data on your own machine. ☁️ Hosted — deploy it remotely for convenient access from anywhere. The idea is to give the user a choice between privacy and convenience.

Built With The current stack includes: React Node.js + Express MongoDB AI/LLM integrations Local AI components Speech-to-text / voice interaction One of the biggest challenges isn't making an AI answer questions—it's connecting AI safely to structured business data.

My goal is to eventually follow: Understand → Preview → Confirm → Execute → Log rather than letting an LLM blindly modify the database.

What I Learned Dukaandaar taught me that adding AI is easy; making AI useful and trustworthy is much harder. Starting with a real person's workflow helped me focus on actual problems instead of adding AI just because it's trendy.

I'm still working on better voice interaction, AI-powered actions, document ingestion, offline-first functionality, self-hosting, and local-to-cloud synchronization. Dukaandaar started with one person's workflow. Now I'm exploring how far that idea can go.

Try It 🔗 GitHub: https://github.com/devrittik/dukaandaar-ai 🔗 Live: https://dukaandaar-ai.vercel.app

Feedback on the AI interaction, privacy model, or the overall idea is always welcome.