← All workin-progress

Riddhi Siddhi

An AI assistant that actually places the order

Role

Design and build, personal project

Timeline

2026

Status

In active development

Stack

FlutterDartGemmaTensorFlow LiteTool callingSupabase
2
apps on one backend
local / cloud
configurable inference

A commerce platform built as two Flutter applications against a single backend: a customer app for browsing and ordering, and a business app where staff run the catalog and fulfil orders.

An assistant with hands

The assistant is not a search box with a chat skin. It calls tools that change real state: looking up products, adding to the cart, and placing orders. When a query matches more than one product it asks which one you meant instead of guessing, and quantity is confirmed before anything reaches the cart.

Inference is configurable. Gemma runs on-device through TensorFlow Lite, or the assistant can call out to a cloud model instead, which matters for a shop where connectivity is uneven and the assistant still has to work.

List scanning reads a handwritten or printed shopping list from the camera and adds the recognized products straight to the cart, which is closer to how people actually shop than typing items one at a time.

The cold-start catalog problem

Most commerce apps demand a complete catalog before launch, which is exactly the work a small business has no time for. This one lets customers order items that are not listed yet. Those requests surface in the business app, and staff add products to the catalog as demand reveals what belongs there.

The business side

  • Categories, brands, products and offer banners, with optional per-product barcodes.
  • Staff accounts and management.
  • Order handling and customer records.
  • Sales insights for the owner.

The app has cleared Google Play review and is in internal testing ahead of public release.

Next systemGemMate