DukaanFlow AI: Making Daily Tiffin Order Management Simpler

Small tiffin and home-food businesses often receive customer requests through chat messages. A customer might ask to skip tomorrow's delivery, change an order, or clarify a subscription. Handling these messages manually can become confusing.

I built DukaanFlow AI – Order Desk to explore how local AI can help sellers organize these requests while keeping important decisions under the seller's control.

What does it do?

DukaanFlow AI is a local-first order desk built with Python, Flask, and SQLite.

Its goal is to help sellers interpret customer messages and review order-related actions through a dashboard.

Some of the key ideas behind the project are:

  • Hinglish message classification: Interpret everyday customer messages, including Roman Hindi and Hinglish.
  • Local AI support: Optionally use Ollama with Qwen2.5 1.5B for structured classification.
  • Rule-based fallback: Use conservative local rules when the AI model is unavailable.
  • Seller confirmation: A one-day tiffin delivery skip must be confirmed by the seller before records are changed.
  • Subscription safety: The system must not automatically cancel a recurring subscription just because a customer asks to skip one delivery.

Why local-first AI?

I wanted to explore an approach that can run locally rather than requiring every message to be sent to a hosted AI service.

The app uses a local Flask application and SQLite database. Ollama can provide additional language interpretation, while a rule-based fallback helps when the model is unavailable.

This is an MVP, not a claim that every Hindi or Hinglish message can already be understood perfectly. Ambiguous requests need conservative handling and seller review.

Tech stack

  • Python
  • Flask
  • SQLite
  • Ollama
  • Qwen2.5 1.5B
  • HTML, CSS, and JavaScript
  • Pytest

Testing the project

I ran the project's test suite locally:

19 tests passed.

These tests are a useful milestone, but they do not mean the application has been validated with real businesses or production traffic.

WhatsApp integration status

The project includes a path for WhatsApp Cloud API integration. However, real WhatsApp traffic is not live by default. It requires the appropriate Meta configuration and credentials before it can be used.

I don't want to claim a production integration that hasn't been configured and tested.

Try the code

The source code is available on GitHub:

DukaanFlow AI – GitHub Repository

You can explore the README for setup instructions, install the Python dependencies, configure your local environment, and run the app on your own machine.

What I learned

This project reminded me that adding AI is only one part of building a useful product. Safe defaults, clear confirmation steps, a fallback when the model is unavailable, and honest integration-status reporting matter just as much.

For small food businesses, even a simple tool that helps organize incoming requests could be useful. The next step is to gather feedback from real sellers and improve the workflow based on their needs.

I'm sharing this project as part of my open-source learning journey. Suggestions, issues, and contributions are welcome!

AI #Python #OpenSource #Hacktoberfest