DukaanFlow AI: A Local-First AI Order Desk for Tiffin Sellers

Description and use

DukaanFlow AI is a local‑first AI order desk for small tiffin services that interprets customer requests via WhatsApp or phone, proposes structured actions, and requires seller confirmation before applying changes.

Problem

Small tiffin services and home‑food businesses receive customer requests via WhatsApp and phone calls; manual handling of daily changes can cause mistakes, especially when customers modify or cancel parts of a subscription.

Outcome

The prototype demonstrates a local‑first AI order desk that reduces manual order‑management work for small food sellers, with a live demo and open‑source code available.

Workflow

• Receive customer request via WhatsApp or phone
• Parse request with local AI (Ollama + Qwen2.5 1.5B)
• Apply rule‑based fallback if AI confidence is low
• Generate proposed order action
• Present action to seller for confirmation
• Apply confirmed changes to order system

Evidence

DukaanFlow AI explores how AI can help sellers interpret requests such as: "Please skip tomorrow's lunch."

Source

The goal is to help turn natural-language messages into structured order actions while keeping the seller in control. Important changes should be reviewed before they are applied.

Source

DukaanFlow AI explores a local-first AI architecture using: Ollama to run a language model locally, Qwen2.5 1.5B for local natural-language interpretation, Rule-based fallback handling, Seller confirmation before important order changes, GitHub Pages to deploy the public demo website.

Source