The product goal
The platform needed to turn WhatsApp into a structured customer-engagement channel without losing the speed and familiarity of a normal conversation.
Each business also needed its own messaging environment on shared infrastructure, so the product was designed multi-tenant from the schema up, bringing CRM, campaigns, events, and automation into one system.
It also needed an operator-facing product around the conversation layer: authentication, role-based access, templates, analytics, payments, rate limiting, and administrative controls.
Platform architecture
- Next.js and TypeScript application layer over a 52-table PostgreSQL database, modelled through Prisma, with Supabase and Redis.
- Meta WhatsApp Cloud API integration with HMAC-validated webhooks, message templates, interactive replies, campaign automation, and live-chat handoff.
- Role-based access control for administrative and operational workflows, scoped per tenant.
- Rate limiting, spam detection, analytics, and Moyasar payment flows.
AI concierge
The concierge combined Gemini with search grounding and vision input. Requests were classified into more than forty intent categories so the system could select an appropriate structured flow instead of treating every message as an open-ended prompt.
The assistant handles multiple languages, designed together rather than bolted on, including right-to-left interface behaviour and location-aware recommendations.
Risk controls
- Webhook signatures are validated before events enter the application workflow.
- Rate limits and spam checks protect the conversation layer from uncontrolled traffic.
- Structured flows handle deterministic actions; model responses are used where interpretation adds value.
- Administrative permissions are separated through role-based access control.
Result
The result was a working end-to-end product rather than a standalone chatbot: inbound WhatsApp events, application logic, AI interpretation, multilingual user experience, administration, and operational safeguards were designed as one system.
It was validated across a 75-user closed beta, and the paid customer-acquisition campaigns it supported increased new-customer acquisition by 12%.
Technology
- Next.js
- TypeScript
- PostgreSQL
- Prisma
- Supabase
- Redis
- Meta Cloud API
- Google Gemini
- Vision AI
- Moyasar
- Multi-tenant SaaS
Last updated