andre rianda

Round 2 of real phone call comparison between two AI voice stacks. The STT sometimes waited 2+ s to decide a sentence had ended. Turns now end on 300 ms of quiet audio plus a transcript that has stopped changing.

Two AI agents just had a 2 minute phone call, real numbers, real phone network, cost per minute (provider prices × this calls actual usage, incl ~$0.01 phone line): modular $0.033 native $0.059

Been building a horror game with Astra using Three.js. Hows the atmosphere looking so far?

Another three.js made by Opus

Tell me “I love you” without telling me “I love you” Opus:

I told Opus 5.5: "put a pianist in the lobby. He has to actually play, not lip-sync." It wired real sheet music to his fingers. 88 keys move, 98% of notes land under the finger that plays them, and everyone in the building hears the same bar at the same second. In a browser 🎹

Opus 5.5 motion design