Raabta AI
WhatsApp automation grounded in your own data
Most support automation fails the same way: it answers confidently and it answers wrong. Raabta AI starts from the opposite end — every reply is retrieved from a knowledge base the business controls, so the model is quoting their material rather than improvising around it.
Teams connect their sources, watch them get chunked and indexed, and then work incoming conversations from a shared inbox. When a thread needs a human, it routes to one. When it needs an appointment, it books one. Usage, delivery and spend are visible on the same dashboard rather than buried in a provider console.
It is built as a single Next.js application over a reactive backend, which means the inbox, the usage meters and the knowledge index all update live without a refresh button anywhere in the product.
- Role
- Product design, full-stack engineering, infrastructure
- Year
- 2026
- Stack
- Next.js
- React
- TypeScript
- Convex
- Clerk
- Twilio
- AI SDK
- Tailwind CSS
What shipped
- Knowledge ingestion from documents, pasted text and crawled pages, with per-source indexing status
- Retrieval-augmented replies grounded in the customer's own material
- Shared team inbox with routing, status controls and full conversation history
- Appointment booking driven straight out of a conversation
- Usage metering, plan limits and billing visibility per workspace
- Admin tooling for managing workspaces, users and entitlements