Inspiration
For decades, shopkeepers, gatekeepers and bus conductors in India counted with a ₹50 metal hand tally counter: one click per item. It could only count. It forgot everything at night and never told you anything.
In my hometown, Karimnagar (Telangana), most small shops still sell without bills. Every night the owner counts cash and UPI with no idea what the total should have been. I wanted to bring the clicker back to life, with a memory, a voice and an AI that explains the day in Telugu.
What it does
One big button per item (Tea, Coffee, Samosa…): one tap per sale, with undo. Expected cash updates live. At closing time the owner enters the UPI total and the cash in the box, and the app shows the gap: money missing, sales not clicked, or all good. Telugu and English, works offline, data stays on the phone.
How we built it
Entirely in Kiro IDE. Three steering files (product, tech, structure) hold the rules every session must follow, above all: money math is tested code; AI never does the arithmetic. From one prompt, Kiro's agent generated the app (about 1,300 lines: logic, UI, Telugu/English strings, service worker) and 46 unit tests, ran them with my approval and fixed until all passed. Deployed on GitHub Pages.
Challenges
Getting money logic right for a real shop. Reviewing Kiro's output, I found that editing an item's price later silently rewrote the money for sales already clicked, which matters to a shop owner. The fix goes through a Kiro spec (requirements → design → tasks).
What we learned
Steering is the difference between "AI wrote some code" and "AI followed my product's rules". Writing down the non-negotiables once made every Kiro session safer. And a human review still catches what tests don't test.
What's next
Speak your sales in Telugu ("rendu tea, oka samosa"), a 3-line AI explanation of the day, and an ESP32 counter button: the physical clicker, truly back from the dead.
Built With
- amazon-bedrock
- aws-lambda
- css3
- esp32
- github
- html5
- javascript
- kiro
- python
- web-speech-api
