Speech to Speech LLM models are great (simple, fast and natural). In this demo this week I used Gemini Live and still the tool calling was pretty reliable and was easy to add to Gemini the missing features I needed: WebRTC and MCP. Not sure why Google doesn't provide those OOTB

My Voice AI Framework Awards 😅 🏆Developer Experience: LiveKit (cleanest api, simplest development and deployment) 🏆Ecosystem: Pipecat (largest community, multi provider, libraries and tools) 🏆More advanced: TEN Framework (multi-language, multi-process, custom VAD)

Trying to do something fun for Gemini x Pipecat Virtual Hackathon in the last minute. Only 4h for the deadline, ooops...

Like many others, Vercel is seeing impact from the AWS outage in us-east-1 (IAD). Our service is global in nature, which means the impact from this outage has varied by customer and region. We’ve taken steps to fail over a number of services. The IAD CDN region was rerouted,

Haven't been able to deploy to Vercel for some hours :(

Quick post after yesterday's AWS outage describing how to design your WebRTC architecture so that it keeps working when an AWS region goes down! For those who miss when I was not talking about AI stuff :) "WebRTC Architecture that keeps working when us-east-1 goes down" Link👇

What is the "best" python SIP stack nowadays? Something simple and well maintained and ideally pure python and not a wrapper of some C library. Maybe Saúl Ibarra Corretgé o Dan Jenkins know?

"Bad architecture is a lot of work that doesn’t change much" newsletter.pragmaticengineer.com/p/what-is-good…