LOCAL-FIRST · OPEN-WEIGHT AI

Less order chaos.
More time for good food.

A lightweight order desk concept for tiffin and home-food sellers. Understand customer requests, review suggested actions, and keep the seller in control.

Try interactive demo ↓ Explore code ↗

✓ Local Ollama option　 ✓ Seller confirmation　 ✓ Rule-based fallback

● ● ● DUKAANFLOW / ORDER DESK ● LOCAL DEMO
MONDAY · ORDER OVERVIEW

Good morning, seller 👋

Requests today12Needs review
Ready to review03Seller action

Recent requests Illustrative preview

🍱
Skip tomorrow's tiffin“Kal ka lunch mat bhejna please”
🔁
Subscription question“Is my monthly plan still active?”

Preview only — no customer data or order records are connected.

Local-firstOptional local inference
Human-in-the-loopSeller confirms order changes
19 tests passedLocal test suite milestone

THE WORKFLOW

AI can suggest.
The seller decides.

Designed for everyday messages in Hinglish, Roman Hindi, and English.

01

Read the request

A seller enters or receives a customer message that needs sorting.

02

Classify locally

Ollama + Qwen2.5 1.5B can interpret supported messages; conservative rules provide a fallback.

03

Review & confirm

The seller checks the proposed action before an order record is changed.

INTERACTIVE WEBSITE DEMO

Try a sample
customer message.

This demo uses simple browser-side rules. It does not call the Python backend or a real AI model.

ILLUSTRATIVE RESULT
🍱
Suggested category

Possible one-day delivery skip

Seller confirmation is required before changing an order record.

Safety rule

A one-day skip must not cancel a recurring subscription.

No data is sent anywhere by this static demo.

WHY OPEN INNOVATION?

Own the workflow.
Choose the model.

Open-weight models and local inference give small projects more control over where classification runs and which model they experiment with. Local execution can reduce hosted API usage, but it still has hardware, electricity, and setup costs.

DukaanFlow AI is an MVP, not a claim that a local model always beats a closed API. Real seller testing and comparative benchmarks are still needed.

PythonFlaskSQLiteOllamaQwen2.5 1.5B
🛡️

SAFE DEFAULTS

One message shouldn't cancel a whole subscription.

Ambiguous requests need review. One-day delivery skips require seller confirmation, and recurring subscriptions should never be automatically cancelled just because someone asks to skip one delivery.

OPEN SOURCE · EARLY MVP

Want to build on it?

Explore the code, review the tests, or suggest what a small food seller would need next.

Open GitHub repository ↗