Key2 - AI-Powered Telecom Platform
Full AI voice layer added to an existing Twilio telecom stack
Visit Key2 CommunicationsAutomated
Call Handling
Deepgram nova-2
STT Model
Sub-500ms
Response Latency
Twilio + OpenAI + ElevenLabs
Integrations
The Problem
Key2 Communications is a Canadian telecom provider serving SMEs. They had an existing Twilio infrastructure for call routing and wanted to add AI-powered automated call handling — without touching or replacing the infrastructure their business already ran on. Manual call routing was the operational bottleneck: common customer queries were going to human agents that an AI could handle, and there was no way to triage calls before a person picked up.
Our Approach
Layered an AI voice system on top of their existing Twilio SIP stack without modifying call routing infrastructure. Incoming calls are intercepted by a FastAPI inference service via Twilio Media Streams. Deepgram nova-2 handles real-time speech-to-text. OpenAI processes the conversation with customer context pulled from Django. ElevenLabs synthesizes the response and streams audio back to the caller sub-500ms. The escalation path routes to human agents with a conversation summary attached so callers never repeat themselves. Django manages business logic and customer state; React surfaces call monitoring and analytics.
Pipeline Breakdown
01 · Collect
- Inbound call capture via Twilio Media Streams
- Real-time STT with Deepgram nova-2
- Customer context pulled from Django CRM
02 · Process
- OpenAI conversation engine with rolling context
- Intent detection and escalation routing
- Sentiment scoring per call turn
03 · Act
- ElevenLabs TTS voice responses streamed back
- Warm transfer to human agents with summary
- Call analytics and monitoring dashboard
Have a similar problem? Let's talk.
← Back to all work