Case Study 03 / AI Agents · Voice · Automation
Voice Agents
Autonomous agents that answer client inquiries and book calls end to end — parsing messages, drafting replies and scheduling, with speech on both ends.
- Role
- AI & automation — team project
- Year
- 2025
- Stack
- n8n, OpenAI GPT, Google Calendar, STT/TTS
- Status
- Shipped
( 01 ) The problem
Answering and booking is constant, manual back-and-forth.
Client inquiries and call booking eat time: reading messages across channels, writing replies, checking a calendar, proposing times, confirming. It's repetitive work that still has to feel personal — and never drop a lead.
( 02 ) The system
One workflow in n8n, from first message to booked call.
A message arrives from any channel — Gmail, Notion or Telegram — and a multi-step workflow parses it, generates a reply with an OpenAI GPT model, and books the call through the Google Calendar API. Every step is wired together with webhooks.
( 03 ) How it works
Parse → generate → schedule, across every channel.
01
Parse
Reads the incoming message and pulls out intent, context and any proposed times.
02
Generate
Drafts a natural, on-brand reply with an OpenAI GPT model.
03
Schedule
Checks availability and books the call through the Google Calendar API.
04
Connect
The same loop runs across Gmail, Notion and Telegram — one workflow, many channels.
( 04 ) Voice
Not just text — a real conversation.
Speech-to-text and text-to-speech sit on top of the workflow, so the agents can hold real customer-service conversations by voice — the caller speaks, the agent listens, answers and books, out loud.
( 05 ) Stack
Glued together with n8n and webhooks.
( 06 ) Outcome
Inquiries answered, calls booked — on their own.
Inquiries get answered and calls get booked around the clock, without someone watching an inbox.
Every message gets a reply and a clear path to a booking, across every channel.
Natural replies and real voice — a conversation, not a rigid menu of options.