Cardynal
AI-native customer support platform
Co-founder & CTO. An intent router arbitrates between deterministic flows and free agents. Trilingual FR/EN/HE with RTL.
1,300 conv./day · ~4s · −80% tickets
- AI agents
- Customer support
- CTO
- Multilingual
The problem
SMBs drown in support requests: rising costs, slow replies, overwhelmed teams. Classic chatbots disappoint because they guess instead of guaranteeing.
The approach
I built an intent router that arbitrates, on every message, between deterministic flows (a state machine) for critical cases and free agents for ambiguity. FR/EN/HE i18n with RTL, WhatsApp, Gmail, Shopify and web-widget integrations, plus a voice sub-application.
The logic that has to be exact stays sovereign; the LLM is only called where judgment would hesitate.
The result
~1,300 support conversations handled a day, ~4-second responses, ~80% fewer tickets. In production, every day, across three languages.