Car dealership accounts kept from a Telegram message

The dealership's books start as a cash book: short lines in Tamil-English shorthand with a voucher number, VARA or PATRU, an amount, a customer and often a vehicle plate.

Copying those lines into ledgers by hand was slow, and a single mistyped figure could leave the books out of balance without anyone noticing.

AI does one job only: reading the shorthand into structured vouchers. Every rule that protects the books, from known account names and dates to both sides balancing and duplicate vouchers, is plain code that runs before anything is saved.

The ledgers stay in Google Sheets, which the owners already used, so nothing about how they read their accounts had to change.

Staff send an entry to the bot and get it back as a clear voucher with Confirm and Cancel buttons. Confirmed lines post to the Day Book and to each account's ledger, and the reply shows the new balances and whether the trial balance still agrees.

If posting fails halfway, every line already written is rolled back. Mistakes are corrected with REVERSE and the voucher number, which builds the opposite entry instead of editing history.

Three more n8n workflows run on their own: a month-end statement snapshot, a daily 02:00 backup of every ledger tab, and an alert to the developer on Telegram if any workflow errors. Every step is written to an audit log tab.