Skip to Main Content

Sisi

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

Sisi — Demo video

Sisi is a WhatsApp-first AI bookkeeper for South African spaza shops, the township corner stores that keep communities fed but still run on handwritten books, memory and trust. Owners message Sisi the way they already talk: text, voice notes and photos, in English, isiZulu, Afrikaans, isiXhosa or mixed slang. Sisi logs sales, reads a photo of a handwritten book page and asks Yes/Undo before saving, keeps credit customers with per-customer statement cards and PDFs carrying the shop's logo, drafts gentle payment reminders (never auto-sent), and sends a daily 7pm nudge, a daily sales PDF and a Sunday snapshot. Tap menus and buttons work, free typing still works, and one message can chain several actions ("sold 2 bread and send Thandi her statement"). All AI runs on the Meta Model API: Muse Spark for understanding, tool planning and reading book photos, and Muse voice transcription for voice notes. Built on WhatsApp, Modal and Turso, with 1,100+ automated tests. No new app, just the chat they already open every day. New: full Sesotho and Setswana support, for six languages in total: English, isiZulu, Afrikaans, isiXhosa, Sesotho and Setswana. Sisi now forecasts stock run-outs from 14-day sales speed, shows them in the 19:00 nudge, and sizes restock lists to match. A "funding pack" command builds a branded PDF (shop profile, 3-month P&L, cash flow, sales trend, credit-book health) for banks or the free Spaza Shop Support Fund. Sisi now sends a branded overview image of the shop's standing after every task, plus native WhatsApp reply buttons and list menus in all 6 languages.

Team