How the AI Calling Pipeline Works
A transparent look into our real-time media bridge, conversational AI state machine, and carrier telephony integration.
Call Initiation & Safety Gate Verification
Whether triggered manually via Quick AI Call, scheduled via an Excel Campaign, or received from an inbound caller, the call passes through our 15-item Telephony Safety Gate. The platform verifies account status, active subscription limits, E.164 phone number formatting, and carrier health diagnostics before dispatching any signal.
Carrier Connect & Dual Stream Setup
The telephony carrier (Plivo or Vobiz) answers the request. The server emits canonical Call Control XML establishing two concurrent operations:
<Record recordSession="true"> captures the complete audio interaction for legal compliance, while <Stream> opens a bidirectional WebSocket connection to the Node.js Telephony Media Bridge.
Real-Time Audio Transcoding & AI Synthesis
PSTN telephone networks transmit audio at 8kHz μ-law (ITU-T G.711). Modern AI speech engines like ElevenLabs require 16kHz PCM audio. Our Node.js media bridge converts incoming μ-law frames into 16kHz linear audio for speech recognition, and downsamples ElevenLabs synthetic voice back into μ-law in under 48ms without dropping a single byte.
Post-Call Analysis, CRM Sync & Private Storage
When the call terminates, the carrier delivers the final recording file via an async callback. The post-call processor transcribes the conversation, evaluates customer qualification, generates an AI summary, and updates your CRM pipeline. The audio file is secured in tenant-isolated private disk storage.