moranai

MoranBuy is an AI reverse marketplace on WhatsApp, built as a product of Moran AI and tested in Tanzania. Instead of searching shops, a customer tells MoranBuy what they want, what they can pay and where they are, by text, by sending a photo of the product, or by a Swahili voice note. For example: “Nina elfu tisini, nataka blenda, nipo Sinza Mori, niletewe”. MoranBuy finds offers from local businesses and ranks them budget first, then by total price including delivery, then by distance. When few local shops fit, Meta’s web search finds the product in shops online near the customer, in their own country and currency, shown as “found online, not confirmed”. It never invents a price or a discount, and it says honestly when nothing fits. The problem: in markets like Kariakoo, more than 200,000 people a day walk shop to shop comparing prices, with more than TSh 10 billion changing hands in cash, and buying from a stranger online is risky. People fear paying for something that never arrives, or arrives wrong. MoranBuy fixes both: One message replaces the walk. Any phone with WhatsApp works. There’s no app to install; you find it by the username @moranbuy, and you can write, speak or send a photo in Swahili or English. Payment protection. The customer pays MoranBuy, which holds the money. The customer gets a receipt PDF with a signed QR code and a private handover code the shop never sees. The shop names a rider, who must send a photo at the door and enter the customer’s code. Only then is the money released to the shop. A wrong or damaged item is refused at the door and refunded. For businesses: list products from one shelf photo, receive orders that are already paid, answer them with one word on WhatsApp, and see what people nearby are asking for. Moran AI businesses not yet on MoranBuy are asked for their consent before any of their products are shown. Every model MoranBuy calls is a Meta model: Muse Spark 1.3 reads requests and photos and writes the replies, Muse Spark web search finds products online, Muse Spark 1.2 audio and Muse Voice Transcribe hear voice notes, SAM 3.1 lists products from a shelf photo and checks delivery photos, and Meta’s MMS voices speak the reply back in Swahili. Fixed rules decide every price, ranking and payment; the models never do. Judges can run the whole story at buy.moranai.co.tz/demo, with no login and no real money: a sandbox shows the customer, the shop owner and the rider on a phone, next to a live server log from the same code that runs in production, with every Meta model call visible. The repo runs the same sandbox from a fresh clone in five minutes (README, first section).