WhatsApp Commerce Suite

Kenyan businesses live on WhatsApp, but pointing a language model at a customer and letting it run the whole order is how you invent a menu item or confirm a total that does not match the cart. These three needed the opposite: an order flow that is correct every time, not plausible most of the time.

1. 01A deterministic state-machine conversation engine: every state, transition and reply explicitly defined. The bot cannot say anything it was not built to say. The natural-language layer in Wanjii Bot sits on top of this same core.
2. 02Full ordering flow, greet by name, recognise returning customers, browse categories to items to variants, multi-item cart, delivery address and phone capture, order summary, confirmation, payment instructions.
3. 03Per-user conversation state persisted in PostgreSQL, so a customer can walk away mid-order and come back to it.
4. 04An admin API for order management and payment verification, with a Next.js dashboard on top.
5. 05Hotel Feedback Pulse on the same base: short guest prompts during a stay, follow-up capture on anything neutral or negative, sentiment and category tagging, and a real-time staff alert when something goes wrong, a multi-hotel data model from the start.
6. 06Re-seeded for Jaba Juice retail as a separate service, so two businesses could evolve without sharing a codebase.

An architecture proven across three businesses in two sectors, and the foundation the productised Wanjii Bot was built on.