Kairo AI Receptionist

About the Project

Kairo deploys persona-specific AI voice and chat receptionists built on a two-portal Next.js TypeScript architecture, a client configuration portal and an admin management dashboard, with Supabase as the auth and conversation data layer. n8n orchestrates the multi-step AI workflow pipelines; Twilio handles inbound voice and WhatsApp channels; Stripe powers subscription billing. Cloudinary manages knowledge-base media; Cloudflare provides CDN and security. Nodemailer delivers conversation summaries and escalation alerts; Google Analytics and Google Search Console drive inbound discovery for AI receptionist queries.

How It Works

1. 1

   The two-portal Next.js monorepo (client portal + admin dashboard) is deployed behind Cloudflare; Supabase stores client configurations, conversation logs, and subscription state under row-level security policies.

2. 2

   n8n orchestrates the full AI workflow: knowledge base ingestion → vector embedding → RAG retrieval → LLM inference → TTS synthesis → Twilio voice delivery, all wired as a visual no-code workflow triggerable via webhook.

3. 3

   Each AI receptionist persona is bootstrapped with a domain-specific knowledge base uploaded via Cloudinary and loaded into a vector store; the RAG retriever reranks chunks by semantic proximity at inference time.

4. 4

   Twilio Voice handles inbound call routing with a streaming STT → LLM → TTS pipeline delivering AI responses in under 800ms; escalation logic summarises the conversation and warm-transfers to a human when uncertainty exceeds a confidence threshold.

5. 5

   Stripe handles tiered SaaS billing with webhook-driven entitlement updates in Supabase; Nodemailer sends post-call summaries, escalation alerts, and weekly conversation analytics to client inboxes.

Tech Stack