Tin Chu
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Families in Colombia overpay for groceries every week because prices for the same product vary a lot between D1, Jumbo, Éxito and Carulla, and nobody has time to compare them. Rinde is a shopping agent that fixes that. You type your list, paste it, or photograph a handwritten one. Rinde finds each item at every store, checks prices live, and gives you a plan, for example "D1 for these 12 items, Jumbo for these 3, and you save $75.930." It also works out whether an extra stop is worth it on foot, by bus, moto or car, and hands you the route in Google Maps or Waze. If you're over budget, Rinde suggests another store, another brand or a smaller pack, each with its exact saving, and nothing changes until you accept. Photographed receipts and DIAN e-invoices feed fresh in-store prices back into the app. **How Meta Muse is used:** Muse reads list photos and receipts, works out what the shopper means by "leche" or "arroz," and decides whether two listings with different barcodes are the same product using names and package photos, with a one-line reason. In a held-out test of 95 products it picked the wrong product 0 times. Muse never produces or touches a price: every price comes from deterministic code with its store, source and time, and every step falls back to a rule if the model is unavailable.