A phone call. An AI conversation.
A working telephone prototype connecting Asterisk to OpenAI Realtime. Call a number, speak and hear the AI respond over the same call.
- Role
- Prototype development & integration
- Product
- Voice helpdesk research / AgeVolt
- Period
- 2026
Overview
I developed an AI voice-assistant prototype for AgeVolt, connecting telephone calls through Asterisk to OpenAI Realtime via a separate Node.js and TypeScript service. The prototype supported spoken conversations over a phone call. Company knowledge and task-specific instructions were the next stage of development.
My work
- Built the Node.js/TypeScript middleware connecting Asterisk call events and audio to OpenAI Realtime.
- Worked on call handling, bidirectional audio and Slovak-language responses.
- Tested real phone conversations, turn-taking and interruption of the assistant’s response.
- Coordinated with O2 on SIP connectivity and investigated trunk registration and deployment issues.
What I worked on
Asterisk to OpenAI
Asterisk handles the telephone side; a standalone Node.js service bridges call events and audio with OpenAI Realtime through WebSocket connections. The prototype separates telephony from the AI conversation layer.
A conversation over the phone
Worked on session setup, audio buffering, end-of-speech detection and interruption when the caller starts speaking. Testing focused on making a spoken exchange work through a telephone call.
Prototype validated; knowledge next
The working result was a callable voice prototype. Connecting company knowledge and refining domain instructions were the planned next steps. Production telephony rollout was a separate stage.
Result
A functional development prototype that could be called for a spoken conversation with an AI assistant. Knowledge-base integration remained the next step.
Technologies
- Asterisk / SIP
- OpenAI Realtime API
- Node.js
- TypeScript
- Express
- Zod
- WebSocket
- Docker / Compose