An AI That Answers Every Call, Text and Email
A receptionist that picks up in real time across voice, SMS and email, captures the caller's details, books the appointment, and hands off to a human the moment it should — with a full contact-centre toolset behind it.
Industry
SaaS / AI Communications
Solution
AI Receptionist & CCaaS
Engagement
~10 Weeks Hand-Built
Services
AI & Full-Stack Development
The Calls Nobody Was There to Answer
For a small business, a missed call is not an inconvenience — it is the lead going to whoever picks up next. Staff are with a customer, or it is after hours, and the phone rings out. Meanwhile the enquiry that did land in the inbox sits there for a day.
The channels made it worse: voice, SMS and email each handled separately, none of them sharing what the customer had already said. And the off-the-shelf voice bots that promised to fix it had a three-to-six-second pause before every reply — long enough that callers assumed the line had dropped.
The Problems We Set Out to Solve
The intake is explicit that this problem list is inferred from the product and its docs rather than stated by the client, so we present it as the product's design brief — which is exactly what it is.
Missed inbound calls turning into lost leads, after hours and at busy times
No instant follow-up on a missed call or a new enquiry
Voice, SMS and email handled separately, with no shared customer memory
Off-the-shelf voice bots too slow and too robotic to keep a caller on the line
No lightweight control tower for owners to see and reschedule follow-ups
One Agent, Every Channel, Sub-Second
We built an AI receptionist that answers on any channel, remembers the customer across all of them, and chases the follow-up itself — on an architecture with no monolithic server anywhere in it.
Streaming Voice, Not Turn-Taking
A websocket relay streams the language model's tokens straight to the telephony layer as they are generated, instead of waiting for the whole reply. Internal measurement: time-to-first-word around 0.6 seconds, down from three to six.
Unified Customer Memory
Voice, SMS and email write to the same customer record, so identity, intent and bookings carry across channels instead of each one starting from nothing.
A Follow-Up Engine That Does Not Forget
A three-touch sequence — SMS, then call, then email — with reschedule, reactivation, per-channel retries and cancellation notices, running without anyone chasing it.
Event-Driven, Not a Monolith
Around ninety serverless edge functions do the work, with multi-tenant isolation enforced in Postgres row-level security scoped by workspace.
What the Platform Does
AI Voice Receptionist
Real-time inbound-call conversation that captures the caller's details and produces a transcript and summary of every call.
SMS AI Agent
A per-number agent that answers SMS threads in the same voice and with the same memory as the phone agent.
Email AI Agent
Inbound and outbound AI email conversation and intake, feeding the same customer record.
Follow-Up Engine
Three-touch reminders with reschedule, reactivation, per-channel retries and cancellation notifications.
Live Human Transfer
Smart transfer to a person — on request, or when the agent detects the caller is getting frustrated.
Switchboard / Console
A browser softphone and call-centre console for the humans behind the AI.
Handover Queue
Contact-centre routing, queues, agent sessions and dispositions.
Numbers Management
Search, purchase and price phone numbers, with per-number routing rules.
Customer Inbox
A 360 view per customer — history, notes, per-customer memory, and handoff by link or PDF.
Agent Builder & Voice Studio
Configure the agents and the voices they speak with, without touching code.
Flows
A visual flow editor for building the conversation paths the agents follow.
Billing
Checkout, invoices and usage rating tied to what the workspace actually consumed.
Growth Suite
Content and SEO generation for blogs and resource pages, with sitemap and llms.txt output.
Partner Portals
Separate dashboards for partner, affiliate and contributor personas.
QA & Analytics
QA scorecards, call analysis and confidence scoring on what the agent actually said.
Serverless, Streaming, Multi-Tenant
Frontend
Backend
Data & Tenancy
AI & Telephony
Built Live, Against a Ringing Phone
- 1
Bootstrap & Scaffold
The first shape of the product, stood up fast so there was something concrete to react to.
- 2
Core Voice + SMS MVP
The AI receptionist answering real calls and real texts, verified live rather than in a demo.
- 3
Omni-Channel Expansion
The email agent, the unified customer memory across all three channels, and the follow-up engine.
- 4
Backend Migration
Moved off the low-code scaffold platform onto a self-owned Postgres project and our own deployment — the client owns the backend outright.
- 5
Voice Latency Overhaul
Replaced turn-taking with a streaming relay. Internal measurement: time-to-first-word fell from three-to-six seconds to around 0.6.
- 6
CCaaS Tooling
The switchboard, handover queue, dispositions and opt-out handling that turn an AI agent into a contact centre.
What Was Delivered
~255
Application Routes
~90
Edge Functions
~141
DB Migrations
~8
Major Integrations
Every inbound call, text and email is answered — in business hours or at two in the morning.
Internal measurement: voice time-to-first-word around 0.6 seconds, down from three to six. Measured in-house, not independently benchmarked.
A customer who called yesterday and texts today is the same customer to the agent — one memory across voice, SMS and email.
The backend runs on infrastructure the client owns, migrated off the low-code platform it was born on.
Frequently Asked Questions
Yes — that latency was the core problem we set out to solve here. We replaced turn-taking with a websocket relay that streams the language model's tokens straight to the telephony layer as they are generated, rather than waiting for the whole reply. Our internal measurement put time-to-first-word at around 0.6 seconds, down from the three-to-six-second pause that makes callers think the line has dropped. That figure is measured in-house, not an independent benchmark.
Services Behind This Build
Explore Other Case Studies
Losing leads to a phone nobody answers?
We build AI agents that pick up on every channel, remember the customer, and hand off to a human at exactly the right moment.
Talk to a Delivery Expert