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

Tadasouk turns a phone video into an online store, for small merchants in Morocco who sell through social media posts today. The problem: these shops, often run from home, sell on Instagram and WhatsApp. A store builder asks them for product photos, written descriptions, prices typed one by one, stock numbers and a company. Most give up and keep posting videos that get "Price?" in the comments. What the merchant does: open the app (a PWA, nothing to install), press the red record button, show each product to the camera and talk in Darija, French, Arabic or English: "this one is 120 dirhams, 20 percent off, I have 6". Or upload a sales video they already posted. What the app does with Meta's models: 1. A detector on the phone outlines the product being held. When it stays still for 800 ms, or the merchant says "hada", SAM 3.1 cuts it out and the phone puts it on a white background. 2. Muse Voice Transcribe 1.0 turns each short stretch of speech into text, and Muse Spark 1.3 turns that text into actions on the product card: price, discount, name, stock, remove, undo. When a command is unclear, the card shows a question mark and the caption asks one question. 3. For each product, Muse Spark 1.3 writes the name and description in French and Arabic, files it in a category and checks prices in Moroccan shops with web search (lowest, average, highest, suggested price). Muse Image 1.0 makes extra photos from other angles, starting from the real photo. 4. The products go to a public store at tadasouk.com/{store}: categories, search, product pages, cart, and a checkout with cash on delivery or credit. A WhatsApp share sends a preview with the photo and the price. The back office also listens. The merchant says what to change (a price, putting a product online, a category, sold out, delivery, a promotion code) and the store does it, answers aloud, and keeps an undo. Other tools: stock count from a shelf photo, a stock camera fixed over the counter, a credit book filled by voice ("Ahmed took two bottles of oil for 240 dirhams, he will pay Friday") with WhatsApp reminders, a counter sale with barcode scanning, and ad texts in French, Darija and Arabic. Built during the event: the repository started empty on 2026-10-27. Next.js on Cloudflare Workers, Supabase Postgres with row level security, Cloudflare R2 for photos. The Meta API key stays on the server; each call is charged to the merchant's AI credit, checked against a daily cap and logged. Not part of this submission: card payments use two existing external services, Stripe (card processing) and Tchentchen (a payment service built before the hackathon). Neither one's code is in the repository, and card payment stays switched off in production. Instagram account import waits for Meta App Review, so the demo uses video upload from the phone. Try it: https://tadasouk.com (sign up, then film a few products). Demo store: https://tadasouk.com/test