Built this for the YC Voice Agents hackathon the other day. Most voice agents are 1:1 — I wanted a hub-and-spoke model instead: everyone talks to the agent concurrently, and it decides per utterance who needs to hear what and conveys it to the other users simultaneously. I hadn't seen that shape before, so the concept seemed worth exploring. This is mainly for coordinating large groups of people over voice, while maintaining a higher signal to noise for what is relevant per user, over standard 'radio' chatter.

Not production-ready, and not easy to set up — it leans heavily on the sponsor stack (Daily, Twilio, Cekura, AWS-hosted model endpoints). So it's a POC, but was a fun build from a system-design and agents angle.

How do you handle interruptions? If a user starts talking while a "spoke" is processing, how does the hub signal it to stop cleanly? Managing the partial state from the interrupted spoke is a problem we've certainly donated a few brain cells to.