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

Across informal retail shops and kiosks in East Africa, counter transactions happen in seconds: customers point at physical items and speak naturally in Kiswahili. Traditional barcode scanners, POS terminals, and manual entry are too slow, expensive, and fragile for these fast-paced environments. SokoSeal Point transforms any standard smartphone camera into an honest, multimodal retail assistant using Meta AI technologies. The shopkeeper points their phone at the counter. As a customer points at items and speaks ("Nipe hii mbili, na hii moja" / "Give me two of this, and one of this"), SokoSeal Point simultaneously segments physical merchandise with Segment Anything (SAM 3.1), tracks pointing dwell order using MediaPipe, and transcribes spoken Kiswahili via Meta Muse Voice. Meta Muse Spark interprets spontaneous natural speech, seamlessly handling mid-sentence self-corrections ("Nipe hii moja, hapana, ziwe tatu...") and items named by voice alone ("...na kikombe moja") without pointing. Spoken quantities and pointed items are catalogue-matched against an on-device inventory with verified reference photos. To eliminate AI hallucinations, deterministic safety prevents adding external or unlisted items (like an external mouse or empty counter space). Once verified, the merchant taps Seal, permanently locking the transaction into an immutable, read-only proof receipt. Featuring instant bilingual localization (English & Kiswahili), SokoSeal Point brings transparency, speed, and trust to everyday retail.