Inspiration
Market vendors keep their books on paper or in their heads. Supplier messages arrive as chat text, sales are scribbled down, and pricing questions have no quick answer — especially when the market has no signal.
What it does
StallSmart is an offline-first AI bookkeeper for market vendors. Vendors paste a supplier message and the app parses it into structured stock — quantities, units and prices — then saves it to Firestore inventory with one tap. Sales are logged in two taps and the dashboard shows live revenue, cost and profit. An Ask box answers real pricing questions.
How we built it
Built with Firebase AI Logic Hybrid Inference: short messages are routed on-device (Gemini Nano) and long documents go to the cloud (gemini-3.5-flash-lite), with an honest ON-DEVICE / IN-CLOUD badge and routing reason on every answer. When the device has no signal, long cloud jobs fail gracefully with a clear needs-internet message instead of a silent failure.
Other Firebase products: Firestore (inventory + sales), Authentication (anonymous), App Check (reCAPTCHA Enterprise), Remote Config, Analytics. The app works offline-first and is free to use — judges can try it live with no sign-up.
Built with Kiro (the coding agent) plus a small manual fix. Note: in the demo recording browser the on-device model was unavailable, so the video honestly shows the cloud fallback path with its fallback note — the hybrid routing code is real and switches to on-device whenever Nano is available.
Challenges we ran into
Making hybrid routing honest: the app must show where each answer really ran, and degrade gracefully offline instead of failing silently.
What's next
More supplier-message formats, multi-stall support, and richer pricing insights for vendors.
Built With
- firebase-ai-logic
- firebase-app-check
- firestore
- gemini
- javascript
- kiro
- vite