Cardynal Voice
Real-time AI phone agents
An AI phone line built solo: multilingual speech-to-speech (FR/EN/HE) on Pipecat + Twilio.
- Voice
- Twilio
- Real-time
- Multi-tenant SaaS
The problem
A missed call is often a lost customer — and staffing a phone line is expensive, especially across several languages.
The approach
A multi-tenant SaaS built solo: real-time speech-to-speech (OpenAI Realtime / Gemini Live) on Pipecat + Twilio, with a multi-provider cascade fallback. AI-generated onboarding + an in-browser WebRTC voice test before going live, per-call cost/latency observability, and Hebrew-specific engineering.
The result
The agent picks up when the client doesn't, qualifies the call, collects the details and sends an SMS summary. Multilingual FR/EN/HE, in real use — for SMBs and an Israeli home-care association.