Full-duplex voice
Speech to LLM to speech, with barge-in.
Real-time voice agent
Full-duplex speech to LLM to speech.
voiceagent
01 / THE OVERVIEW
Speech in, speech out, without waiting.
The voice agent runs full-duplex speech to LLM to speech with interruption handling. It captures CRM data during the call and sends call activity to an analytics dashboard.
“Listen, think, and speak at the same time.”THE IDEA
02 / THE PROBLEM
Turn-taking breaks in real calls.
People interrupt. They pause. They change their mind. A voice agent that waits for a clean sentence feels slow and unnatural.
03 / THE SOLUTION
Interruptions are part of the flow.
Built on Node.js and WebSocket with Deepgram and Groq, the agent handles interruptions, captures CRM data, and reports call activity. The demo reported sub-200 ms latency in demo conditions.
CRM capture
Call details saved during the conversation.
Analytics
A dashboard for call activity and outcomes.
04 / THE IMPACTReported
Reported
sub-200 ms.
Reported sub-200 ms latency in demo conditions. The same system was adapted into AI receptionists for dental clinics.
View the voice agent on GitHub
The cover is a real screenshot of the live voice-agent call screen.
NEXT IN THE COLLECTION