Baba Time
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Baba Time lets a small shop owner run their in-store screens by voice. Independent shops (bakeries, cafes, grocers) own screens but rarely use them: making a promo visual takes time, design skills and a computer. With Baba Time, the owner holds a button on their phone and says "20% off croissants until 11am, on the window screen". Seconds later, the window TV shows a professional promo visual, and it disappears on its own at 11:00. How it works: 1. The voice command is transcribed. 2. An agent understands the intent (product, discount, end time, target screen) and orchestrates the steps through tool calls. 3. The product photo taken on the phone is cut out. 4. A background is generated, with no text in it. 5. The visual is composed by code: product name, promo price, crossed-out price, end time. 6. The agent inspects its own visual, locates problems (unreadable price, weak contrast), fixes the layout or regenerates the background, then verifies again before publishing. 7. The visual is pushed live to the right screen and expires automatically. Two design choices make it safe for real shops: 1. Prices are never written or calculated by a model. They come from the shop catalog and are rendered by code, so what the screen shows is always exact. 2. The crossed-out price follows the EU Omnibus Directive: it is the lowest price of the last 30 days, not an arbitrary one. The agent refuses a promo that would be misleading. Privacy: voice audio is processed in memory and never stored, photo metadata is stripped, and every screen shows an "AI-generated visual" notice. The demo uses a fictional bakery and its catalog.