Moosa Memon
← Index of work

An overflow receptionist that only answers what the humans missed

RingRelay: a compliance-first AI receptionist for spas, salons, clinics and studios. It rings the staff first, picks up only the calls, texts and emails nobody answered, handles routine questions from a grounded knowledge base, and escalates the moment anything is sensitive.

Type
Agents
Stack
Twilio / LangGraph / FastAPI / n8n / RAG
Status
Complete; escalation rules unit-tested as a pure function

Problem

A business that already has front-desk staff it trusts doesn’t need them replaced. It needs the gaps covered: the call that rang out at lunch, the text that came in after hours, the email nobody saw until Monday. An AI-first front desk is the wrong product for them. An overflow layer is the right one.

System

Twilio rings the humans for 18 seconds. On no answer, busy, or out-of-hours, RingRelay answers. Every call opens with a disclosure that it’s an automated assistant, then runs deterministic escalation rules before any knowledge answer: asked for a person, any health mention, a complaint or refund, anything already escalated. Escalation tries a transfer, then takes a message if that rings out. Every call, without exception, ends with the transcript and recording delivered to staff.

Worth knowing

The escalation rules are graph edges, not prompt instructions: a pure function over call state that decides the route before the model gets to phrase anything. “Any health mention” escalates in a clinic whether or not the model thought it was handling it fine. That’s the difference between something a clinic owner can deploy and something they’d be nervous about, and it’s also what makes the behaviour testable without a model in the loop.

Have a workflow that looks like this?

Most of these start as a messy, manual process someone got tired of. Twenty minutes is usually enough to sketch how I'd approach yours.