Kairo AI Receptionist

The Problem

Small and mid-sized service businesses, dental clinics, salons, real estate agencies, law firms, lose an estimated 30–45% of inbound leads to missed calls and after-hours inquiries. Hiring a full-time receptionist costs ₹18,000–₹35,000/month and still doesn't cover weekends or peak hour overflow.

Generic chatbots trained on nothing industry-specific frustrate callers with irrelevant responses. What was needed was an agent that understood the vocabulary of each industry, had authority to book appointments, and knew when to get a human involved.

Constraints

• Intent classification had to complete in < 80 ms, any slower and the conversation felt laggy.

• Must work across WhatsApp, website chat, and Twilio Voice from a single reasoning engine, no separate codebases.

• Operators are non-technical: the entire deployment and knowledge-base upload flow had to require zero developer involvement.

• Escalation logic had to be configurable per industry without code changes, a law firm and a salon have very different thresholds.

• No hallucinations on appointment booking, incorrect slots or double-bookings would destroy operator trust instantly.

How Nexolve Built It

Kairo is a multi-layer AI system, a fast intent classifier feeds a RAG-augmented generation core, which has live access to the business's calendar API and a configurable escalation engine. Each industry vertical ships with its own seed knowledge base.

1. 1

   Intent classification layer

   Every inbound message is routed through a fast intent classifier (fine-tuned on industry-specific corpora). Classification takes <80ms and determines whether the interaction is an appointment request, pricing inquiry, complaint, or general enquiry before the main model even begins generating.

2. 2

   Industry-trained knowledge base

   Each vertical (dental clinic, salon, real estate agency, law firm) ships with a pre-built knowledge base structured as vector embeddings in pgvector. At deployment, the operator uploads their own FAQs and procedures, which are chunked, embedded, and merged into the shared base.

3. 3

   Appointment booking via calendar API

   When a booking intent is confirmed, Kairo calls the Google Calendar / Calendly API to fetch real-time availability, presents slots to the caller, and creates a confirmed booking, all within the conversation, without human intervention.

4. 4

   Escalation guardrails

   Kairo monitors confidence scores per turn. If the model's certainty drops below a configurable threshold (e.g. complex legal questions, distressed callers), it triggers a warm handoff: SMS + push notification to the human team with full conversation context prepended.

5. 5

   Multi-channel delivery

   The same agent core powers WhatsApp (Twilio), website chat widget, and voice (Twilio Voice + Deepgram STT). Channel adapters handle format translation; the reasoning engine is shared across all surfaces.

Tech Stack

Outcomes

• Zero after-hours leads lost, Kairo handles inbound 24/7 across WhatsApp, chat, and voice.

• Appointment booking fully automated, no staff touch required for standard scheduling flows.

• Human escalation triggered with full context, agents receive complete conversation transcript before joining.

• Deployment time under 48 hours for a new industry vertical using the seed knowledge base + operator's own FAQ upload.

• Live at aireceptionist-6a4.pages.dev, open for demo bookings.