MLQU 009 · Early build
Munshi
Indian shopkeepers speak, and the quotes, reminders, and bills come out.
The problem
A shopkeeper will not type into a business app. They will talk. Every tool built for them assumes a keyboard, English, and an hour to learn the interface, so none of them get used.
How it’s built
Talk to it
A Flutter shell with four tabs where the primary input is speech in the owner’s own language rather than a form.
Orchestrate
A FastAPI backend turns an instruction into the right action: a quotation, a reply, a payment reminder, a reorder, or a bill.
Business Brain
A Postgres model layer holding what the shop actually is, so the assistant answers from the business rather than from the last message.
Swappable everything
Speech, OCR, payments, WhatsApp and accounting all sit behind provider interfaces chosen by environment variable. Changing a vendor does not touch orchestration.
Run it cheaply
Cloud Run in asia-south1 with keyless deploys through Workload Identity Federation, against Supabase Postgres.
What changed
- The first set of flows works end to end
- Switching a speech or payment vendor is one environment variable
- Built for the phone in the shop, not for a desktop nobody there uses
Want one of these?
Two lines is enough. We reply within a day.